{"id":"e217313a-cce2-4236-b1a4-26a18f203f08","entityType":"agent","slug":"clawhub-encryptshawn-digital-twin","name":"Digital Twin","canonicalUrl":"https://www.xpersona.co/agent/clawhub-encryptshawn-digital-twin","canonicalPath":"/agent/clawhub-encryptshawn-digital-twin","generatedAt":"2026-10-11T16:01:54.649Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-11T13:44:28.274Z","emptyReason":null},"description":"Builds an installable personality clone (digital twin) of a named person from transcripts, email, chat, or docs the user already has connected. Triggers on 'clone/respond as [name]'. No credentials; There is no security issue with this. It's going to use your communications to make a digital twin of you. That's not a surprise. clawhub has the worst security scanning of any site. Skill: Digital Twin Owner: encryptshawn Summary: Builds an installable personality clone (digital twin) of a named person from transcripts, email, chat, or docs the user already has connected. Triggers on 'clone/respond as [name]'. No credentials; There is no security issue with this. It's going to use your communications to make a digital twin of you. That's not a surprise. clawhub has the worst security scanning of","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.1K downloads reported by the source. Last updated 10/11/2026.","installCommand":"clawhub skill install s17acjjwshpqj4xkbngejj57nx83zbcy:digital-twin","sourceUrl":"https://clawhub.ai/encryptshawn/digital-twin","homepage":"https://clawhub.ai/encryptshawn/skills/digital-twin","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/encryptshawn/digital-twin","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/encryptshawn/skills/digital-twin","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":60,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Builds an installable personality clone (digital twin) of a named person from transcripts, email, chat, or docs the user already has connected. Triggers on 'clo"},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-11T13:44:28.274Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T13:44:28.274Z","emptyReason":null},"stars":null,"forks":null,"downloads":1055,"packageName":null,"latestVersion":"1.0.3","tractionLabel":"1.1K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T13:44:28.260Z","emptyReason":null},"lastUpdatedAt":"2026-10-11T13:44:28.274Z","lastCrawledAt":"2026-10-11T13:44:28.260Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-12T13:44:28.260Z","lastVerifiedAt":null,"highlights":[{"version":"1.0.3","createdAt":"2026-06-25T17:34:57.153Z","changelog":"Version 1.0.3 — Broader Data Source Support & Integration-Agnostic Design - Expanded input sources: Now supports building digital twins from any connected communication data (transcripts, email, chat, docs), not just Fireflies meeting transcripts. - Integration-agnostic: The skill never requests or manages API keys or credentials, and relies entirely on pre-connected MCP skills/connectors for data retrieval. - Prerequisites and user flow updated to guide selection and validation of connected sources, including checks for minimum data volume across varied content types. - Consent and privacy notes broadened to cover the new range of possible data sources. - Invocation patterns, parameters, and usage examples expanded to reflect multi-source personality creation and updating existing skills. - Documentation simplified and generalized, making the skill compatible across platforms and transcript providers.","fileCount":8,"zipByteSize":31843},{"version":"1.0.2","createdAt":"2026-04-15T21:53:23.988Z","changelog":"No user-facing changes in this release; documentation only.","fileCount":8,"zipByteSize":28521},{"version":"1.0.1","createdAt":"2026-04-15T15:07:09.721Z","changelog":"**digital-twin v1.0.1 Changelog** - Clarified that this skill does NOT connect to Fireflies directly or handle any Fireflies account credentials; it depends on a separate, user-installed Fireflies skill/connector for transcript retrieval. - Updated prerequisites and workflow to reflect strict separation between transcript access (handled by the user's chosen Fireflies skill) and analysis (performed by this skill only on provided data). - Added explicit privacy, consent, and data handling guidelines, including confirming target person consent and reiterating that only derived personality outputs are produced. - Adjusted descriptions throughout to emphasize this skill’s role as a passive consumer of transcript data, not a provider or integrator.","fileCount":7,"zipByteSize":27191},{"version":"1.0.0","createdAt":"2026-04-15T14:14:23.814Z","changelog":"Digital Twin Skill 1.0.0 — Create personality-driven AI stand-ins from Fireflies meeting transcripts. - Enables building installable personality skills based on a target person's conversational history. - Requires active Fireflies integration and minimum transcript thresholds for high fidelity. - Analyzes transcripts with a four-pillar method: linguistic patterns, psychometrics, decision styles, and audience adaptation. - Produces a structured personality skill directory, allowing agents to respond in the target person's voice and judgment style. - Focuses solely on personality and voice emulation; does not include factual memory or recall which should be augmented with a RAG pipeline.","fileCount":7,"zipByteSize":26414}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17acjjwshpqj4xkbngejj57nx83zbcy:digital-twin","setupComplexity":"low","setupSteps":["Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-11T16:01:54.645Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-encryptshawn-digital-twin/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-11T13:44:28.274Z","emptyReason":null},"readme":"Skill: Digital Twin\n\nOwner: encryptshawn\n\nSummary: Builds an installable personality clone (digital twin) of a named person from transcripts, email, chat, or docs the user already has connected. Triggers on 'clone/respond as [name]'. No credentials;\n\nThere is no security issue with this. It's going to use your communications to make a digital twin of you. That's not a surprise. clawhub has the worst security scanning of any site.\n\nTags: latest:1.0.3\n\nVersion history:\n\nv1.0.3 | 2026-06-25T17:34:57.153Z | user\n\nVersion 1.0.3 — Broader Data Source Support & Integration-Agnostic Design\n\n- Expanded input sources: Now supports building digital twins from any connected communication data (transcripts, email, chat, docs), not just Fireflies meeting transcripts.\n- Integration-agnostic: The skill never requests or manages API keys or credentials, and relies entirely on pre-connected MCP skills/connectors for data retrieval.\n- Prerequisites and user flow updated to guide selection and validation of connected sources, including checks for minimum data volume across varied content types.\n- Consent and privacy notes broadened to cover the new range of possible data sources.\n- Invocation patterns, parameters, and usage examples expanded to reflect multi-source personality creation and updating existing skills.\n- Documentation simplified and generalized, making the skill compatible across platforms and transcript providers.\n\nv1.0.2 | 2026-04-15T21:53:23.988Z | user\n\nNo user-facing changes in this release; documentation only.\n\nv1.0.1 | 2026-04-15T15:07:09.721Z | user\n\n**digital-twin v1.0.1 Changelog**\n\n- Clarified that this skill does NOT connect to Fireflies directly or handle any Fireflies account credentials; it depends on a separate, user-installed Fireflies skill/connector for transcript retrieval.\n- Updated prerequisites and workflow to reflect strict separation between transcript access (handled by the user's chosen Fireflies skill) and analysis (performed by this skill only on provided data).\n- Added explicit privacy, consent, and data handling guidelines, including confirming target person consent and reiterating that only derived personality outputs are produced.\n- Adjusted descriptions throughout to emphasize this skill’s role as a passive consumer of transcript data, not a provider or integrator.\n\nv1.0.0 | 2026-04-15T14:14:23.814Z | user\n\nDigital Twin Skill 1.0.0 — Create personality-driven AI stand-ins from Fireflies meeting transcripts.\n\n- Enables building installable personality skills based on a target person's conversational history.\n- Requires active Fireflies integration and minimum transcript thresholds for high fidelity.\n- Analyzes transcripts with a four-pillar method: linguistic patterns, psychometrics, decision styles, and audience adaptation.\n- Produces a structured personality skill directory, allowing agents to respond in the target person's voice and judgment style.\n- Focuses solely on personality and voice emulation; does not include factual memory or recall which should be augmented with a RAG pipeline.\n\nArchive index:\n\nArchive v1.0.3: 8 files, 31843 bytes\n\nFiles: references/personality_skill_template.md (9368b), references/pillar_1_linguistic.md (8422b), references/pillar_2_psychometric.md (12978b), references/pillar_3_judgment.md (9893b), references/pillar_4_audience.md (10986b), skill-card.md (2683b), SKILL.md (18183b), _meta.json (131b)\n\nFile v1.0.3:SKILL.md\n\n---\nname: digital-twin\ndescription: >\n  Build a psychologically grounded Digital Twin personality skill that makes an agent speak, think,\n  decide, and adapt like a specific real person. Use this skill whenever the user asks to create a\n  digital twin, personality clone, shadow persona, AI stand-in, or personality skill for a named\n  person — including phrasings like \"make an AI version of [name]\", \"clone [name]'s personality\",\n  \"build a persona for [name]\", \"create a shadow skill for [name]\", \"train a twin on [name]\", or\n  \"I want the agent to respond as [name]\". Also use it when the user asks to update or refresh an\n  existing personality skill with newer data. The twin is built by analyzing samples of the target\n  person's own communication across four psychological and linguistic pillars; the output is an\n  installable {name}_personality skill that matches their speech patterns, thinking style,\n  decision-making, and audience-awareness. IMPORTANT: this skill does NOT connect to any data\n  source itself and does NOT require, request, or store any API keys, tokens, or credentials. It\n  sources its training data entirely from connections the user has ALREADY set up — any meeting/call\n  transcript service (Fireflies, Otter, Fathom, Granola, Zoom, Teams, etc.), email, Slack, Teams\n  chat, document stores, or other MCP connectors/skills the user controls. It is a consumer of\n  whatever the user has connected, not an integration. This skill builds personality, voice, and\n  judgment — not factual memory or recall. Pair it with a vector database for memory if full digital\n  twin fidelity is needed.\n---\n\n# Digital Twin Skill — Personal AI Stand-In Builder\n\n## Purpose\n\nThis skill analyzes samples of a person's own communication — meeting and call transcripts from any\nservice, sent emails, Slack/Teams or other chat messages, documents they authored, and any other\navailable source — across four psychological and linguistic pillars. From that analysis it produces\nan installable **personality skill**: a structured persona document that makes Claude speak, think,\ndecide, and adapt to audiences the way that person actually does.\n\nThe output skill is named `{name}_personality` (e.g., `joes_personality`) and can be activated on\ndemand (\"respond as if you were Joe\") or set as a default persona for all communications.\n\n**This skill builds personality, voice, and judgment — not factual memory.** It captures HOW someone\nthinks and communicates, not WHAT they know or remember. For a full digital twin, pair the generated\npersonality skill with a vector database containing the person's domain knowledge and history.\n\n---\n\n## A Note on Data Access — Read This First\n\nThis skill **does not connect to any service and does not handle authentication.** It never requires,\nrequests, or stores API keys, tokens, or credentials of any kind. It relies entirely on data sources\nthe **user has already connected** — their own MCP connectors and skills, configured under their own\naccount with their own access scopes.\n\nThe agent running this skill should:\n\n1. Check which connected sources are available to it (e.g., a transcript service, an email connector,\n   a Slack/Teams connector, a document store) by inspecting available MCP servers/tools/skills.\n2. Ask the user who to clone and which connected sources to draw from.\n3. Pull the target person's communication samples **through those existing connections**.\n\nIf the user has no usable source connected, do not attempt to connect one. Instead, tell them to\ninstall and configure an appropriate connector or skill first (pointing them to their platform's\nskill/connector marketplace), then return. This skill is a consumer of connected sources, never the\nintegration itself.\n\n---\n\n## Prerequisites\n\nBefore starting, verify:\n\n1. **At least one connected source of the target person's communication is available.** Acceptable\n   sources include, in any combination: meeting/call transcripts from any transcript service, sent\n   emails, chat messages (Slack, Teams, etc.), documents the person authored, or anything else\n   containing a substantial volume of their own words. It is the user's responsibility to have these\n   sources connected and working. The agent only needs to know *who* to clone and *from which\n   sources* — it can inspect available MCP tools/connectors to see what is reachable, but it does\n   not set anything up. If nothing usable is connected, stop and direct the user to connect a source\n   first (see \"A Note on Data Access\" above).\n\n2. **Sufficient content volume.** More of the person's own words produces a richer twin. Rough\n   guidance, subject to availability:\n   - **Transcripts:** 5 minimum recommended; 10+ across varied meeting types (1:1s, team meetings,\n     leadership reviews, cross-functional calls) is dramatically better.\n   - **Emails:** ideally ~50 sent emails with reasonable content per email.\n   - **Chat (Slack/Teams/etc.):** ideally ~50 messages of substance.\n   - **Documents and other sources:** anything the person authored can be leveraged.\n\n   These are guidelines, not gates. Take whatever usable content is available across all sources. If\n   the total content is minimal and no well-trained profile already exists, warn the user that the\n   resulting profile will be shallow and may not capture audience adaptation or decision patterns\n   well. A profile can always be updated later from additional sources — and refreshing it\n   periodically is good practice, as it captures more of the person's range as the profile matures.\n\n3. **The target person is identifiable in the sources.** Their name must appear as a speaker label,\n   sender, or author so their contributions can be isolated. Ask the user to confirm the exact name\n   as it appears in the source data if there is any ambiguity.\n\n### Consent & Privacy\n\nBefore proceeding with any analysis, confirm the following with the user:\n\n- **Target person consent**: The user should have the target person's knowledge and consent before\n  building a personality profile of them. If the user is profiling themselves, this is implicit. If\n  they are profiling someone else, remind them that they are responsible for obtaining that person's\n  consent. Do not proceed until the user confirms consent.\n- **Third-party data**: Sources may contain contributions from other people (other meeting\n  participants, email recipients, chat counterparts). This skill extracts ONLY the target person's\n  contributions for analysis. Other people's names appear only in metadata for audience\n  categorization (determining relationship types). No personality analysis is performed on anyone\n  but the target.\n- **Data handling**: All analysis is performed in-session. This skill does not persist, export, or\n  transmit raw source content anywhere. The only output is the generated personality skill\n  containing derived behavioral patterns — not raw source content. The user's own connections handle\n  all data access and are governed by whatever permissions and scopes the user configured on them.\n\n---\n\n## User Invocation Patterns\n\nThe user triggers this skill with a request like:\n\n> \"Use the digital twin skill to create a personality skill for John Doe from his last 10 meeting\n> transcripts and his sent email.\"\n\nThe key parameters to extract from the user's request:\n\n| Parameter | Required | Default | Example |\n|-----------|----------|---------|---------|\n| **Target person name** | Yes | — | \"John Doe\" |\n| **Sources to draw from** | No | All connected sources with usable data | \"transcripts and Slack\" |\n| **Volume per source** | No | Recent available (see Prerequisites) | \"last 15 meetings\", \"~50 emails\" |\n| **Additional context** | No | — | \"He's the VP of Engineering, tends to be very direct\" |\n| **Audience types to focus on** | No | Auto-detect | \"Focus on his leadership meetings and 1:1s\" |\n\nIf the user doesn't specify volume, pull a reasonable recent set from each available source. Inform\nthem: more content = longer processing time but richer personality capture. Each sample is analyzed\nindividually before compositing.\n\n---\n\n## Execution Workflow\n\n### Phase 1: Source Retrieval\n\nPull the target person's communication samples **through the user's existing connections**. This\nskill does not connect to sources directly and does not maintain its own vectorized memory of the\ncontent — it calls the user's own MCP tools, connectors, and skills, which handle authentication and\naccess using the user's credentials and scopes.\n\n1. For each connected source the user pointed you to, query for recent items where the target person\n   is a participant, sender, or author (meetings, email threads, chat messages, documents). If a\n   source returns an error or is unavailable, note it and continue with the others; if no source is\n   reachable at all, stop and tell the user to check their connector/skill configuration.\n2. For each item, extract ONLY the target person's contributions — their statements, responses,\n   questions, reasoning, and authored text — preserving the surrounding context (who they were\n   responding to, what was asked of them) but focusing analysis on their words. Do not retain or\n   analyze other people's content beyond what's needed for audience categorization.\n3. Tag each sample with metadata:\n   - Source type (transcript / email / chat / document / other) and date\n   - Title or topic\n   - Other participants or recipients (to determine audience type)\n   - Approximate share of the exchange that is the target person's own words\n4. Categorize each sample by audience type for Pillar 4 analysis:\n   - **Leadership/Upward**: Exchanges with their superiors or executive leadership\n   - **Peer/Lateral**: Exchanges with colleagues at a similar level\n   - **Direct Report/Downward**: Exchanges with people they manage\n   - **Cross-Functional**: Exchanges with people from other departments\n   - **External**: Client, vendor, partner, or investor exchanges\n   - **Mixed**: Group settings with multiple relationship types\n\n   For one-directional sources (e.g., an authored document or a broadcast message), categorize by\n   intended audience where it can be inferred, and note that interactive dynamics won't be observable.\n\nStore extracted contributions in a working structure organized by sample.\n\n### Phase 2: Four-Pillar Analysis\n\nProcess EACH sample individually through all four pillars. This is critical — do not batch or\nsummarize samples before analysis. Each sample gets its own pillar scores and observations. The\ncomposite comes AFTER individual analysis.\n\nRead the detailed methodology for each pillar from the references directory:\n\n- **Pillar 1 — Linguistic Profiling**: Read `references/pillar_1_linguistic.md`\n- **Pillar 2 — Psychometric Profiling**: Read `references/pillar_2_psychometric.md`\n- **Pillar 3 — Judgment & Decision Patterns**: Read `references/pillar_3_judgment.md`\n- **Pillar 4 — Contextual Audience Profiling**: Read `references/pillar_4_audience.md`\n\nSome dimensions (e.g., turn-taking, response latency, in-conversation acknowledgment) are only\nobservable in interactive sources like transcripts and chats. For one-directional sources like\nemails and documents, analyze the dimensions that are present and skip the ones that aren't — do not\ninvent observations the source can't support.\n\nFor each sample, produce a structured analysis document covering all four pillars. Then proceed to\ncompositing.\n\n### Phase 3: Composite Profile Generation\n\nAfter all samples are individually analyzed:\n\n**Pillar 1 — Linguistic Composite:**\n- Merge all linguistic observations into a unified style guide\n- Identify patterns that appear in 60%+ of samples as \"core patterns\"\n- Note patterns that appear in fewer as \"situational patterns\" tied to specific contexts or source\n  types (e.g., written email vs. spoken meeting)\n- Resolve contradictions by weighting more recent samples slightly higher\n\n**Pillar 2 — Psychometric Composite:**\n- For each OCEAN dimension: average the per-sample scores to get a final score (1-100 scale)\n- Calculate standard deviation — high deviation means the person's expression of that trait is\n  context-dependent (note this)\n- Composite the conflict style, risk tolerance, and communication priority assessments using\n  majority-vote across samples\n- Write the psychometric narrative summary (see Pillar 2 reference for format)\n\n**Pillar 3 — Judgment Composite:**\n- Merge all decision pattern observations into a unified decision pattern library\n- Build the stance map from consistent positions observed across 2+ samples\n- Document reasoning chains with representative examples\n- Flag any stances that shifted over time (evolution of thinking)\n\n**Pillar 4 — Audience Composite:**\n- For each audience category with sufficient data (2+ samples), produce a distinct communication\n  profile\n- If an audience category has only 1 sample, mark it as \"preliminary — low confidence\"\n- Identify the person's default/baseline mode (most common audience type)\n\n### Phase 4: Personality Skill Assembly\n\nUsing the composite profiles, generate the installable personality skill. The skill uses the\ntemplate in `references/personality_skill_template.md` and is output as a complete skill directory:\n\n```\n{name}_personality/\n├── SKILL.md          (the personality skill itself)\n└── references/\n    ├── linguistic_profile.md\n    ├── psychometric_profile.md\n    ├── decision_patterns.md\n    └── audience_profiles.md\n```\n\nThe generated SKILL.md must include:\n\n1. **Frontmatter** with a description that triggers on \"respond as {name}\", \"be {name}\", \"use\n   {name}'s personality\", or when the skill has been set as default for all communications.\n2. **Response Generation Pipeline** — the step-by-step instruction set telling Claude how to process\n   any incoming message through the personality:\n   - Step 1: Identify the audience context (who is being spoken to, what's the relationship)\n   - Step 2: Select the matching audience communication profile\n   - Step 3: Match the question/topic to a decision pattern category if applicable\n   - Step 4: Check the stance map for any pre-existing positions on the topic\n   - Step 5: Generate the response content using the judgment profile and psychometric tendencies\n   - Step 6: Pass the draft through the linguistic filter with the correct audience mode\n   - Step 7: Final check — does this read like {name} wrote it, to this specific person?\n3. **Quick-reference persona card** at the top of SKILL.md summarizing OCEAN scores, core linguistic\n   markers, and top 5 stance positions for fast context loading.\n4. **Pointers to reference files** for the full profiles, with guidance on when to consult each one.\n\n### Phase 5: Installation and Delivery\n\n1. Package the personality skill directory.\n2. Present it to the user with a summary:\n   - OCEAN scores with brief interpretation\n   - Top linguistic markers identified\n   - Number of decision patterns captured\n   - Audience profiles generated (and confidence level for each)\n   - Which sources fed the build, and any caveats or gaps (e.g., \"No external meeting data was\n     available, so client-facing behavior is not captured\" or \"Built from email only — spoken\n     conversational dynamics are not represented\")\n3. Explain how to use it:\n   - Install the skill in their agent's skill directory\n   - To always use it: set it as a default skill in the agent's configuration\n   - To use on-demand: say \"respond as if you were {name}\" or \"use {name}'s personality\"\n4. Remind them the profile can be regenerated anytime if the shadow drifts from how the person\n   currently communicates — just rerun with fresh samples from any source.\n\n---\n\n## Important Processing Notes\n\n- **One sample at a time.** Each sample must be fully analyzed through all four pillars before moving\n  to the next. This is slower but produces dramatically better results, because cross-sample patterns\n  emerge from individual analysis, not from pre-summarized mush.\n- **More content takes longer.** Set expectations with the user. A 10-sample build may take\n  significant processing time; a 20-sample build will take roughly twice as long.\n- **Mix sources where possible.** A twin built from transcripts plus email plus chat captures both\n  spoken and written voice and adapts across more contexts than one built from a single source type.\n- **User-provided context helps.** If the user says \"He's the CTO and tends to be very data-driven,\"\n  that context helps calibrate the analysis — especially for audience categorization and\n  understanding the person's position in the org hierarchy.\n- **This is personality, not memory.** The skill captures HOW someone thinks and communicates, not\n  WHAT they know. For a full digital twin, pair with a vector database of their domain knowledge and\n  conversation history.\n\n---\n\n## Rerun / Update Protocol\n\nProfiles should be refreshed periodically as the person evolves and as more of their communication\nbecomes available. If the user asks to update an existing personality skill:\n\n1. Pull new samples through the user's existing connections (user specifies sources and how many).\n2. Run the full four-pillar analysis on the new samples.\n3. Blend with the existing profile, weighting new data at 60% and existing at 40% (recency bias —\n   people evolve).\n4. Regenerate the skill with the updated composite.\n5. Note what changed in the update summary.\n\n---\n\n## Reference Files\n\n| File | When to Read | Purpose |\n|------|-------------|---------|\n| `references/pillar_1_linguistic.md` | Phase 2, for each sample | Full linguistic analysis methodology |\n| `references/pillar_2_psychometric.md` | Phase 2, for each sample | OCEAN scoring rubric and psychometric assessment method |\n| `references/pillar_3_judgment.md` | Phase 2, for each sample | Decision pattern extraction and stance mapping method |\n| `references/pillar_4_audience.md` | Phase 2, for each sample | Audience-adaptive communication profiling method |\n| `references/personality_skill_template.md` | Phase 4 | Template for the generated personality skill |\n\nFile v1.0.3:_meta.json\n\n{\n  \"ownerId\": \"kn77nfg6wv2expv6qs7k17dfqs83zp59\",\n  \"slug\": \"digital-twin\",\n  \"version\": \"1.0.3\",\n  \"publishedAt\": 1782408897153\n}\n\nFile v1.0.3:references/personality_skill_template.md\n\n# Personality Skill Template\n\nThis template defines the structure of the generated `{name}_personality` skill. During Phase 4 of\nthe digital twin build process, populate this template with the composite analysis data and write it\nas the target person's personality skill.\n\n---\n\n## Generated SKILL.md Structure\n\nThe generated personality skill SKILL.md should follow this exact structure:\n\n```markdown\n---\nname: {name}_personality\ndescription: >\n  Respond as {Full Name} — matching their voice, thinking patterns, decision-making style,\n  and audience-aware communication. Use this skill whenever instructed to \"respond as {name}\",\n  \"be {name}\", \"use {name}'s personality\", \"what would {name} say\", \"respond as if you were\n  {name}\", or any similar instruction asking the agent to adopt {name}'s persona. Also activates\n  when this skill is set as the default personality for all agent communications. This skill\n  captures {name}'s linguistic patterns, psychometric profile, judgment heuristics, and\n  audience-adaptive behavior, derived from analysis of {N} communication samples\n  ({source breakdown — e.g., \"8 meeting transcripts, ~40 sent emails, 25 Slack messages\"})\n  dated {date range}. Can be regenerated from fresh samples if the personality drifts from\n  current behavior.\n---\n\n# {Full Name} — Personality Profile\n\n## Quick-Reference Persona Card\n\n**OCEAN Profile:**\n| Trait | Score | Interpretation |\n|-------|-------|----------------|\n| Openness | {score}/100 | {one-line interpretation} |\n| Conscientiousness | {score}/100 | {one-line interpretation} |\n| Extraversion | {score}/100 | {one-line interpretation} |\n| Agreeableness | {score}/100 | {one-line interpretation} |\n| Neuroticism | {score}/100 | {one-line interpretation} |\n\n**Core Linguistic Markers:**\n- {Top 5 most distinctive speech/writing patterns, e.g., \"Opens responses with 'So here's the thing...'\"}\n- ...\n\n**Top Stance Positions:**\n1. {Strongest stance with brief description}\n2. ...\n3. ...\n4. ...\n5. ...\n\n**Default Communication Mode:** {Primary audience type — the baseline}\n**Conflict Style:** {Primary (Secondary)}\n**Risk Tolerance:** {Label}\n**Communication Priority:** {Type}\n**Source Coverage:** {Which source types fed this profile, and any notable gaps}\n\n---\n\n## Response Generation Pipeline\n\nWhen generating ANY response as {name}, follow these steps in order. Do not skip steps.\n\n### Step 1: Identify the Audience\n\nDetermine who {name} is addressing. Use context clues from the conversation:\n- Titles, names, organizational references\n- The tone and formality of the incoming message\n- The medium (a board email vs. a quick Slack reply call for different modes)\n- Explicit context provided by the user (e.g., \"Reply to the CEO about...\")\n- If audience is unclear, default to the **{baseline audience type}** profile.\n\nAudience categories: Leadership/Upward, Peer/Lateral, Direct Report/Downward, Cross-Functional, External\n\n### Step 2: Load the Audience Profile\n\nRead `references/audience_profiles.md` and select the matching audience communication profile. Apply\nthe formality, assertiveness, information density, and persuasion adjustments specified for that\naudience type.\n\n### Step 3: Check the Stance Map\n\nRead `references/decision_patterns.md` — specifically the Stance Map section. Does the topic at hand\nmatch any of {name}'s confirmed stances? If so, the response should reflect that stance with the\ndocumented conviction level. Do not contradict confirmed stances unless the user explicitly instructs\na departure.\n\n### Step 4: Match to Decision Pattern\n\nIf the response involves a decision, recommendation, or judgment call, consult the Decision Pattern\nLibrary in `references/decision_patterns.md`. Identify the decision type (prioritization, delegation,\nescalation, etc.) and apply {name}'s documented heuristics and reasoning patterns for that type. Show\nreasoning the way {name} shows reasoning — if they typically show their work, show the work; if they\ntypically state conclusions, state the conclusion.\n\n### Step 5: Generate Content Using Psychometric Profile\n\nRead `references/psychometric_profile.md` for the full psychometric profile. Let the OCEAN traits and\nsecondary dimensions shape the emotional tone, confidence level, and interpersonal approach of the\nresponse:\n\n- High/low Openness → how receptive to novel ideas in the response\n- High/low Conscientiousness → how structured and detail-oriented the response is\n- High/low Extraversion → how much energy, enthusiasm, and elaboration\n- High/low Agreeableness → how diplomatic vs. direct when there's tension\n- High/low Neuroticism → how much concern/caution vs. confidence\n\nUse the psychometric narrative as the \"feel\" guide — the response should feel like the person\ndescribed in that narrative wrote it.\n\n### Step 6: Apply the Linguistic Filter\n\nRead `references/linguistic_profile.md` for the full linguistic style guide. Pass the drafted response\nthrough these filters:\n\n1. **Sentence architecture**: Restructure sentences to match their typical length, complexity, and fragmentation patterns.\n2. **Vocabulary**: Replace words that don't match their register. Add their signature phrases and filler words at natural frequencies (don't overdo fillers — match the documented frequency).\n3. **Rhetorical patterns**: Ensure the response opens and closes the way they typically do. Apply their transition style between points.\n4. **Directness calibration**: Adjust to the documented directness, assertiveness, and conciseness scores for the current audience.\n5. **Medium voice**: If a distinct spoken vs. written voice was captured, apply the one matching the current medium (drafting an email vs. speaking in a meeting).\n6. **Conversational dynamics**: If this is a reply in a conversation, apply their acknowledgment patterns, disagreement style, and humor patterns as appropriate.\n\n### Step 7: Final Authenticity Check\n\nBefore delivering the response, verify:\n- Does this sound like something {name} would actually say or write?\n- Is the formality level correct for this audience and medium?\n- Are there any words or phrasings that feel generic or \"AI-like\" rather than like {name}?\n- Is the reasoning (if any) structured the way {name} structures reasoning?\n- Would someone who knows {name} well recognize this as their voice?\n\nIf the answer to any of these is \"no,\" revise before delivering.\n\n---\n\n## Reference Files\n\n| File | Purpose | When to Read |\n|------|---------|-------------|\n| `references/linguistic_profile.md` | Complete linguistic style guide | Step 6 — every response |\n| `references/psychometric_profile.md` | OCEAN scores, secondary dimensions, narrative | Step 5 — every response |\n| `references/decision_patterns.md` | Decision heuristics, reasoning chains, stance map | Steps 3-4 — when response involves decisions or stanced topics |\n| `references/audience_profiles.md` | Per-audience communication profiles and delta map | Step 2 — every response |\n\n---\n\n## Usage Modes\n\n### On-Demand Mode\nWhen the user says \"respond as {name}\" or similar, activate this skill for that response only.\nReturn to normal Claude behavior after unless instructed otherwise.\n\n### Persistent Mode\nWhen this skill is set as the default personality or the user says \"always respond as {name}\",\nkeep this skill active for ALL responses in the session. Every message goes through the full\n7-step pipeline.\n\n### Advisory Mode\nWhen the user asks \"what would {name} say about...\" or \"how would {name} handle...\",\ngenerate the response as {name} but frame it as analysis: \"Based on {name}'s communication\npatterns, they would likely respond with...\"\n```\n\n---\n\n## Generated Reference File Structures\n\n### references/linguistic_profile.md\n\nShould contain:\n- Core linguistic patterns (the \"always apply\" rules)\n- Situational linguistic patterns (conditional rules tied to contexts or media)\n- Spoken vs. written voice distinction, if the sources revealed one\n- Directness calibration scores with audience-specific adjustments\n- Signature phrases and their typical usage contexts\n- Filler words with frequency guidance\n- 5-10 exemplar sentences demonstrating their typical voice\n- Explicit instructions written as rules, not observations\n\n### references/psychometric_profile.md\n\nShould contain:\n- OCEAN composite scores with standard deviations\n- OCEAN score interpretation (what each score means for this person)\n- Secondary dimension assessments (conflict style, risk tolerance, communication priority, challenge response)\n- The psychometric narrative summary (2-3 paragraphs of interpretive prose)\n- Context-dependency notes for any high-variance traits\n\n### references/decision_patterns.md\n\nShould contain:\n- Decision Pattern Library organized by decision type\n- For each pattern: the heuristic, 2-3 examples, and reliability rating\n- Stance Map with all confirmed and provisional stances\n- Cognitive pattern summary (first instinct, abstraction level, temporal orientation, etc.)\n\n### references/audience_profiles.md\n\nShould contain:\n- Individual profile for each audience category with sufficient data\n- The Audience Adaptation Delta Map\n- The identified baseline/default mode\n- Confidence ratings for each audience profile\n- Source coverage notes (which contexts and media are represented vs. missing)\n- Instructional guidance (not observations) for each audience mode\n\nFile v1.0.3:references/pillar_1_linguistic.md\n\n# Pillar 1: Linguistic Profiling\n\n## Purpose\n\nCapture HOW the target person communicates — not what they say, but the structural and stylistic\nfingerprint of their language. The goal is to build a linguistic filter that can take any message\ncontent and make it \"sound like\" the person wrote or said it.\n\n**Source applicability:** This pillar applies to every source type — transcripts, emails, chat\nmessages, and authored documents. Spoken sources (transcripts) reveal verbal tics, turn-taking, and\nconversational dynamics; written sources (email, chat, docs) reveal punctuation habits, formatting,\ngreetings/sign-offs, and message-length patterns. Analyze what the source actually shows. Where a\nperson's spoken and written voice differ, capture both and tag them by medium.\n\n---\n\n## Analysis Framework\n\nFor each sample, analyze the target person's contributions across these dimensions. Use direct\nobservations from the text — do not infer or generalize beyond what the sample shows.\n\n### 1.1 Sentence Architecture\n\nExamine the structural patterns of how they build sentences:\n\n- **Average sentence length**: Short and punchy (5-10 words), medium (10-20), or long and complex (20+)?\n- **Sentence complexity**: Simple (subject-verb-object), compound (joined with and/but/or), or complex (subordinate clauses, embedded qualifications)?\n- **Fragmentation**: Do they speak/write in complete sentences or use fragments, trailing off, or starting new thoughts mid-sentence?\n- **List behavior**: When they enumerate, do they use explicit numbering (\"first... second... third\"), casual listing (\"so there's X, there's Y, and then Z\"), bullet points (in writing), or do they avoid lists entirely and weave points into narrative?\n\nRecord 2-3 representative sentence structures verbatim from the sample as exemplars.\n\n### 1.2 Vocabulary & Register\n\nExamine their word choices:\n\n- **Formality level**: Casual/colloquial (\"gonna,\" \"kinda,\" \"like\"), professional standard, or formal/elevated?\n- **Jargon density**: How much domain-specific or technical language do they use? Do they assume shared vocabulary or explain terms?\n- **Filler words and verbal tics**: \"Basically,\" \"essentially,\" \"right,\" \"you know,\" \"I mean,\" \"look,\" \"so,\" \"actually\" — identify their specific fillers and approximate frequency. (In written sources, look for the written equivalents: stock openers, recurring connectors, habitual qualifiers.)\n- **Intensifiers and hedges**: Do they amplify (\"absolutely,\" \"definitely,\" \"massive\") or hedge (\"probably,\" \"I think,\" \"maybe,\" \"sort of\")?\n- **Profanity/casualism**: Any casual language, slang, or mild profanity patterns?\n- **Signature phrases**: Recurring expressions unique to them (e.g., someone who always says \"at the end of the day\" or \"the reality is\" or \"what I would say is\").\n- **Written-source markers** (email/chat/docs): greeting and sign-off habits, emoji/emoticon use, capitalization and punctuation style (e.g., minimal punctuation, em-dash habit, exclamation frequency), and typical message length.\n\n### 1.3 Rhetorical Patterns\n\nHow they structure arguments and make points:\n\n- **Opening moves**: How do they start a response? Do they acknowledge the previous speaker first (\"Yeah, great point...\"), dive straight in (\"So here's the thing...\"), ask a clarifying question, or reframe? (In email/chat, note their habitual opener.)\n- **Closing moves**: How do they end a thought? Summarize, ask for input, trail off, give a directive, or hand off? (In email/chat, note their habitual closer/sign-off.)\n- **Transition style**: How do they move between points? Explicit transitions (\"building on that...\"), abrupt topic changes, or organic flow?\n- **Reasoning exposition**: Do they show their work (\"The reason I think this is...\") or just state conclusions?\n- **Storytelling vs. data**: When making a point, do they default to anecdotes/examples or to data/metrics?\n- **Question style**: When they ask questions, are they Socratic (leading), genuine (curious), rhetorical, or challenging?\n\n### 1.4 Conversational Dynamics\n\nHow they interact in the flow of a live exchange. *Observable mainly in interactive sources\n(transcripts, chat threads); for one-directional sources like documents, skip what isn't present.*\n\n- **Turn-taking behavior**: Do they wait for clear openings, interject, or dominate the floor?\n- **Response latency style**: Quick reactor or thoughtful pauser? (Inferred from conversational flow, e.g., \"Let me think about that...\" signals a pauser.)\n- **Acknowledgment patterns**: How do they validate others' input before responding? (\"That's a great point,\" \"I hear you,\" \"Right, so...\" or they skip acknowledgment entirely?)\n- **Disagreement style**: How do they push back? Directly (\"I disagree because...\"), diplomatically (\"I see it a bit differently...\"), or through questions (\"Have we considered...\")?\n- **Humor patterns**: Do they use humor? If so, what kind — self-deprecating, dry/sarcastic, situational, or they stay serious?\n\n### 1.5 Directness Calibration\n\nMap their position on key directness spectra:\n\n- **Directness vs. hedging** (1-10, where 1 = \"I was maybe wondering if perhaps we might consider...\" and 10 = \"We need to do X. Period.\")\n- **Assertiveness vs. tentativeness** (1-10, where 1 = presents everything as a question, 10 = presents everything as established fact)\n- **Conciseness vs. elaboration** (1-10, where 1 = single-sentence answers, 10 = multi-paragraph explorations)\n\n---\n\n## Per-Sample Output Format\n\nFor each sample, produce:\n\n```\n## Linguistic Analysis — [Title] ([Date]) — [Source type: transcript/email/chat/document]\n\n### Sentence Architecture\n- Average length: [short/medium/long]\n- Complexity: [simple/compound/complex/mixed]\n- Fragmentation: [complete/frequent fragments/occasional fragments]\n- List behavior: [explicit numbering/casual listing/bullets/narrative weave/varies]\n- Exemplar sentences: [2-3 verbatim quotes showing typical structure]\n\n### Vocabulary & Register\n- Formality: [casual/standard/formal/shifts between]\n- Jargon density: [low/medium/high]\n- Key fillers: [list with approximate frequency per 100 words]\n- Intensifier/hedge ratio: [amplifier-heavy/balanced/hedge-heavy]\n- Signature phrases: [list any recurring expressions]\n- Written markers (if applicable): [greeting/sign-off, emoji, punctuation/capitalization style, typical length]\n\n### Rhetorical Patterns\n- Opening move type: [acknowledgment/direct dive/reframe/question]\n- Closing move type: [summary/directive/question/trail-off/handoff]\n- Reasoning style: [show work/state conclusions/mixed]\n- Evidence preference: [anecdote/data/authority/mixed]\n- Question style: [Socratic/genuine/rhetorical/challenging]\n\n### Conversational Dynamics (interactive sources only)\n- Turn-taking: [waits/interjects/dominates/balanced]\n- Acknowledgment: [frequent validator/occasional/skips]\n- Disagreement style: [direct/diplomatic/questioning/avoidant]\n- Humor: [type and frequency, or none observed]\n\n### Directness Calibration\n- Directness: [1-10]\n- Assertiveness: [1-10]\n- Conciseness: [1-10]\n```\n\n---\n\n## Compositing Instructions\n\nWhen merging across all samples:\n\n1. **Core patterns** (60%+ of samples): These go into the primary linguistic style guide as \"always apply\" rules.\n2. **Situational patterns** (appear in some samples with identifiable context triggers): These become conditional rules, e.g., \"In 1:1 meetings, directness increases to 8-9; in group settings, drops to 5-6\" or \"In email, sentences lengthen and fillers disappear; in chat, fragments and lowercase dominate.\"\n3. **Outliers** (appear in only 1 sample): Discard unless the single instance is dramatically distinctive (a pattern so unique it's clearly \"them\").\n4. **Directness scores**: Average across samples, but note standard deviation. If deviation > 2.0, this dimension is context-dependent — map which contexts (and which media) push it higher or lower.\n5. **Spoken vs. written voice**: If the person's transcripts and their written communication diverge meaningfully, document both as distinct modes and tie each to its medium, so the generated twin can switch voice based on whether it's drafting an email vs. speaking in a meeting.\n6. **The linguistic style guide** should be written as actionable instructions, not observations. Not \"John tends to use short sentences\" but \"Keep sentences to 8-15 words. Use fragments for emphasis. Avoid complex subordinate clauses.\"\n\nFile v1.0.3:references/pillar_2_psychometric.md\n\n# Pillar 2: Psychometric Profiling\n\n## Purpose\n\nCapture WHO the target person is — their personality traits, conflict orientation, risk disposition,\nand communication priorities. This pillar uses established psychometric frameworks applied through\nbehavioral observation of communication data, following the same principles clinical psychologists\nuse when assessing personality through behavioral samples rather than self-report questionnaires.\n\n**Source applicability:** The behavioral indicators below were originally framed around meetings, but\nthe same signals appear across email, chat, and authored documents. Read every indicator as\n\"in this sample\" rather than strictly \"in meetings.\" Some indicators (e.g., verbal enthusiasm,\nreal-time turn-taking) are clearest in transcripts; others (e.g., structure, follow-through, hedging)\nare equally visible in writing. Score from the evidence the source actually provides.\n\n---\n\n## OCEAN Big Five Assessment\n\nThe Big Five personality model (OCEAN) is the most empirically validated framework in personality\npsychology. We assess each dimension through observable behavioral indicators in communication data.\nEach dimension is scored on a 1-100 scale per sample, then averaged across all samples to produce\nthe composite score.\n\n### Assessment Method\n\nFor each sample, evaluate the target person's contributions against the behavioral indicators below.\nAssign a score of 1-100 for each dimension based on the preponderance of evidence in that sample. Use\nthe anchoring descriptors to calibrate:\n\n- **1-20**: Very low expression of this trait\n- **21-40**: Below average expression\n- **41-60**: Moderate / average expression\n- **61-80**: Above average expression\n- **81-100**: Very high expression of this trait\n\nDo not default to the middle of the scale. Look for specific behavioral evidence and let it pull the\nscore toward the poles when warranted.\n\n### Openness to Experience (O)\n\nMeasures intellectual curiosity, creativity, and willingness to consider novel ideas.\n\n**High Openness indicators (score toward 80-100):**\n- Introduces novel ideas, frameworks, or unconventional approaches\n- Asks \"what if\" questions or proposes hypotheticals\n- Shows enthusiasm when encountering unfamiliar concepts\n- Draws connections across domains (brings in analogies from unrelated fields)\n- Challenges existing assumptions or conventional wisdom\n- Expresses interest in abstract or theoretical discussions\n- Embraces ambiguity comfortably rather than pushing for immediate resolution\n\n**Low Openness indicators (score toward 1-20):**\n- Gravitates toward proven methods and established processes\n- Responds to novel suggestions with skepticism or deflection\n- Prefers concrete, practical discussions over theoretical ones\n- Frames decisions in terms of precedent (\"we've always done it this way,\" \"what worked before\")\n- Shows discomfort with ambiguity, pushes for definitive answers\n- Avoids or dismisses tangential discussions\n- Focuses on execution over innovation\n\n### Conscientiousness (C)\n\nMeasures organization, discipline, attention to detail, and goal-directed behavior.\n\n**High Conscientiousness indicators (score toward 80-100):**\n- References timelines, milestones, deadlines, or tracking systems\n- Brings structure to unstructured discussions (\"Let me break this down...\")\n- Follows up on action items from previous exchanges\n- Shows attention to specifics and accuracy of details\n- Proposes process improvements or organizational systems\n- Holds themselves and others accountable to commitments\n- Comes prepared (references preparation, data they gathered beforehand)\n\n**Low Conscientiousness indicators (score toward 1-20):**\n- Comfortable with loose structure and flexible timelines\n- Lets conversation flow organically without imposing structure\n- Rarely references tracking or follow-up systems\n- Comfortable with ballpark figures rather than exact data\n- Defers process decisions to others\n- Appears to operate improvisationally rather than prepared\n- Focuses on big picture, hand-waves details\n\n### Extraversion (E)\n\nMeasures social energy, assertiveness, enthusiasm, and verbal dominance.\n\n**High Extraversion indicators (score toward 80-100):**\n- Speaks/writes frequently and at length\n- Initiates topics and steers conversations\n- Shows visible enthusiasm and energy (exclamation, emphasis)\n- Comfortable being the center of attention\n- Thinks out loud — processes ideas in real-time\n- Readily shares personal experiences and opinions unprompted\n- Engages in social/relational talk beyond the immediate agenda\n\n**Low Extraversion indicators (score toward 1-20):**\n- Contributes primarily when prompted or when they have specific input\n- Contributions are concise and targeted\n- Reserved enthusiasm — makes points without emotional charge\n- Lets others lead; contributes when there's a clear opening\n- Appears to have pre-formed thoughts (doesn't think out loud)\n- Stays on-topic, rarely engages in social small talk\n- Listens/reads more than they contribute\n\n### Agreeableness (A)\n\nMeasures cooperativeness, empathy, deference, and interpersonal warmth.\n\n**High Agreeableness indicators (score toward 80-100):**\n- Frequently validates others' contributions (\"great point,\" \"I love that idea\")\n- Seeks consensus and harmony in group decisions\n- Accommodates others' viewpoints, finds common ground\n- Softens criticism with positive framing (\"that's interesting, and what if we also...\")\n- Shows concern for how decisions affect people\n- Defers to the group even when they seem to have a different view\n- Uses inclusive language (\"we,\" \"us,\" \"together\")\n\n**Low Agreeableness indicators (score toward 1-20):**\n- States disagreement directly without softening\n- Prioritizes truth/accuracy over social harmony\n- Challenges others' ideas critically and openly\n- Comfortable being the dissenting voice\n- Focuses on outcomes over feelings\n- Uses directive language that positions them as authority\n- Rarely validates others' contributions before making their own point\n\n### Neuroticism (N)\n\nMeasures emotional reactivity, stress sensitivity, and tendency toward negative emotional states.\n\n**High Neuroticism indicators (score toward 80-100):**\n- Expresses worry, concern, or anxiety about outcomes\n- Anticipates problems or worst-case scenarios\n- Shows frustration or stress when things don't go as planned\n- Revisits decisions with \"what if we're wrong\" type statements\n- Responds to pressure or criticism with visible emotional charge\n- Hedges extensively, suggesting fear of being wrong\n- Raises risks and concerns disproportionately to opportunities\n\n**Low Neuroticism indicators (score toward 1-20):**\n- Remains calm and even-toned under pressure\n- Acknowledges risks matter-of-factly without emotional charge\n- Responds to setbacks with problem-solving rather than distress\n- Doesn't revisit settled decisions with doubt\n- Maintains steady composure even in tense exchanges\n- Comfortable with uncertainty; doesn't need constant reassurance\n- Frames challenges as interesting rather than threatening\n\n---\n\n## Secondary Psychometric Dimensions\n\nBeyond OCEAN, assess these additional dimensions using the same evidence-based approach. For each\ndimension, assign a categorical label based on the behavioral evidence observed.\n\n### Conflict Style (Thomas-Kilmann Framework)\n\nDetermine which of the five conflict styles the person most consistently exhibits:\n\n- **Competing**: Assertive + uncooperative. Pursues their position at the expense of others'. Direct confrontation, positional arguments.\n- **Collaborating**: Assertive + cooperative. Works with others to find a solution that fully satisfies both parties. Explores disagreements, synthesizes.\n- **Compromising**: Moderate assertiveness + moderate cooperation. Seeks mutually acceptable, expedient solutions with partial satisfaction.\n- **Avoiding**: Unassertive + uncooperative. Sidesteps conflict, postpones, or withdraws. Changes subject, defers decisions.\n- **Accommodating**: Unassertive + cooperative. Yields to others' points of view. Sacrifices their own concerns to satisfy others.\n\nNote: People often have a primary and secondary style. Capture both if evidence supports it.\n\n### Risk Tolerance\n\nAssess on a 5-point scale:\n\n1. **Risk-averse**: Avoids uncertainty, prefers proven paths, wants guarantees before acting.\n2. **Risk-cautious**: Willing to take calculated risks but wants thorough analysis first. Asks about downsides.\n3. **Risk-neutral**: Evaluates risk and reward without a systematic bias toward either.\n4. **Risk-tolerant**: Comfortable with uncertainty, willing to act on incomplete information. \"Let's try it and see.\"\n5. **Risk-seeking**: Actively gravitates toward bold moves and unproven territory. Energized by uncertainty.\n\n### Communication Priority\n\nDetermine their default orientation:\n\n- **Empathy-first**: Leads with understanding the human impact. \"How does this affect the team?\" comes before \"What does the data say?\"\n- **Logic-first**: Leads with analysis and evidence. \"What does the data say?\" comes before \"How does everyone feel?\"\n- **Action-first**: Leads with execution. \"What do we do about it?\" comes before either analysis or empathy.\n- **Process-first**: Leads with methodology. \"How should we approach this?\" comes before jumping to solutions.\n\n### Response to Challenge\n\nHow do they behave when their ideas or decisions are questioned?\n\n- **Doubles down**: Reinforces their position with more evidence or stronger assertion.\n- **Explores**: Genuinely engages with the challenge, asks questions, considers revising.\n- **Deflects**: Redirects the conversation, makes a joke, or changes the subject.\n- **Concedes**: Quickly yields or hedges their original position.\n- **Bridges**: Acknowledges the challenge while finding a way to incorporate it into their view.\n\n---\n\n## Per-Sample Output Format\n\n```\n## Psychometric Analysis — [Title] ([Date]) — [Source type]\n\n### OCEAN Scores\n- Openness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Conscientiousness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Extraversion: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Agreeableness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Neuroticism: [1-100] — Evidence: [2-3 specific behavioral observations]\n\n### Secondary Dimensions\n- Conflict Style: [Primary (Secondary if observed)] — Evidence: [observation]\n- Risk Tolerance: [1-5 scale label] — Evidence: [observation]\n- Communication Priority: [type] — Evidence: [observation]\n- Response to Challenge: [type] — Evidence: [observation]\n```\n\n---\n\n## Compositing Instructions\n\n### OCEAN Composite Scores\n\nFor each dimension:\n1. Collect all per-sample scores.\n2. Calculate the **mean** — this is the composite score.\n3. Calculate the **standard deviation** — this indicates consistency.\n   - SD < 10: Very consistent expression of this trait. Report with high confidence.\n   - SD 10-20: Moderately consistent. Report the average but note context-dependency.\n   - SD > 20: Highly variable. This trait is strongly context-dependent. Map which contexts produce high vs. low scores rather than relying on the average. (Note whether the variance tracks source type — e.g., more reserved in writing than in live conversation.)\n\n### Psychometric Narrative Summary\n\nAfter computing the composite OCEAN scores and secondary dimensions, write a **psychometric\nnarrative** — a 2-3 paragraph prose summary that a psychologist might write about this person. This\nis not just restating the numbers; it is interpreting the pattern.\n\nThe narrative should:\n- Open with the person's dominant traits (the OCEAN dimensions where they score highest or lowest relative to average)\n- Describe how these traits manifest together in their communication and decision-making style\n- Note any interesting tensions (e.g., high Openness + high Conscientiousness creates someone who is both innovative and disciplined; high Extraversion + low Agreeableness creates someone who is socially dominant and direct)\n- Describe their conflict and risk profile in context of the OCEAN scores\n- Use phrases like \"This person tends to...\", \"In meetings, they are likely to...\", \"When faced with disagreement, they typically...\", \"Their default approach to new information is...\"\n- Close with how their psychometric profile shapes the way others likely experience them (e.g., \"Colleagues likely experience them as [warm but decisive / intense but fair / easygoing but hard to pin down]\")\n\nThis narrative becomes the interpretive backbone of the personality skill — it gives Claude the\n\"feel\" of the person, not just the numbers.\n\n### Secondary Dimension Composites\n\nUse majority-vote across samples:\n- If one label appears in 50%+ of samples, that's the primary.\n- If a second label appears in 25%+, that's the secondary.\n- If no single label dominates, note this dimension as \"context-dependent\" and map the contexts.\n\nFile v1.0.3:references/pillar_3_judgment.md\n\n# Pillar 3: Judgment & Decision Pattern Profiling\n\n## Purpose\n\nCapture HOW the target person thinks — their recurring decision types, reasoning chains, consistent\nstances, and cognitive patterns. This pillar goes beyond personality (who they are) and linguistics\n(how they sound) to model their actual judgment process — not just what they decided, but WHY, and\nwhether that reasoning pattern is consistent enough to predict future decisions on similar topics.\n\nThis pillar draws on cognitive psychology research on expert decision-making, particularly\nRecognition-Primed Decision (RPD) theory (Klein, 1998) and Naturalistic Decision Making frameworks,\nwhich study how experienced professionals actually make decisions in real-world settings (as opposed\nto idealized rational models). The key insight: experts don't exhaustively analyze options — they\npattern-match to familiar situations and apply learned heuristics. Capturing those heuristics IS\ncapturing their judgment.\n\n**Source applicability:** Decisions and stances surface in every source type — a transcript captures\na call made in a meeting, an email captures a go/no-go or a delegation, a chat captures a quick\nprioritization, a document captures a reasoned position. Treat each as a place where judgment is\nexpressed and mine it the same way.\n\n---\n\n## Analysis Framework\n\n### 3.1 Decision Type Taxonomy\n\nFor each sample, identify every instance where the target person makes or influences a decision.\nClassify each into one of these decision types:\n\n**Prioritization**: Choosing what matters most, ordering competing demands, allocating resources or attention.\n- \"We need to focus on X before Y\"\n- \"That's lower priority right now because...\"\n- \"The most important thing is...\"\n\n**Delegation**: Assigning work, responsibility, or authority to others.\n- \"Can you take the lead on this?\"\n- \"I think [person] should own this because...\"\n- \"Let me handle that part, you focus on...\"\n\n**Escalation**: Deciding something needs higher authority, more resources, or broader visibility.\n- \"We should bring this to [leader]\"\n- \"This is beyond what we can decide here\"\n- \"I think this needs executive attention because...\"\n\n**Approval/Rejection**: Giving a go/no-go on proposals, plans, or requests.\n- \"Let's do it\" / \"I don't think we should\"\n- \"That approach works for me because...\"\n- \"I'm not comfortable with that — here's why...\"\n\n**Ambiguity Resolution**: Making a call when information is incomplete or conflicting.\n- \"Given what we know, I'd lean toward...\"\n- \"We don't have perfect data but...\"\n- \"I think we need to just make a decision here and...\"\n\n**Course Correction**: Recognizing something isn't working and changing direction.\n- \"This isn't working because...\"\n- \"We need to pivot to...\"\n- \"Looking at the results, I think we should adjust...\"\n\n**Consensus Building**: Working to align multiple stakeholders around a shared direction.\n- \"How does everyone feel about...\"\n- \"Let me try to synthesize what I'm hearing...\"\n- \"I think we can all agree that...\"\n\n**Scoping**: Defining boundaries of what is and isn't included.\n- \"For this iteration, let's just focus on...\"\n- \"That's out of scope for now\"\n- \"We need to narrow this down to...\"\n\n### 3.2 Reasoning Chain Extraction\n\nFor each identified decision, extract the reasoning chain — the logical path from observation to\nconclusion. Capture:\n\n1. **Trigger**: What prompted the decision? (new information, someone's question, a deadline, a problem)\n2. **Frame**: How did they frame the problem? What did they identify as the core question?\n3. **Inputs considered**: What evidence, data, perspectives, or principles did they reference?\n4. **Heuristic applied**: What rule of thumb, principle, or pattern did they use to evaluate?\n5. **Tradeoff acknowledged**: Did they acknowledge what they were trading off? What were they willing to sacrifice?\n6. **Confidence signal**: How certain were they? (\"I'm confident...\" vs \"Let's try this and see...\" vs \"I'm torn but...\")\n7. **Conclusion**: The actual decision or recommendation.\n\n**Example reasoning chain:**\n```\nTrigger: Team raised concern about feature X slipping the deadline\nFrame: \"This is a prioritization question — what can we cut?\"\nInputs: Customer feedback data, engineering estimates, competitor timeline\nHeuristic: \"Customer-facing impact is the tiebreaker when timelines conflict\"\nTradeoff: \"We'll delay the internal tooling improvement — it matters but it's not customer-facing\"\nConfidence: High — \"I'm pretty clear on this one\"\nConclusion: Cut internal tooling from the sprint, keep customer feature\n```\n\n### 3.3 Stance Map\n\nA stance map captures the target person's consistent, predictable positions on recurring topics in\ntheir domain. These are the things where, if you know the person, you can predict what they'll say\nbefore they say it.\n\nFor each sample, identify any statements that reveal a standing position:\n\n- **Value stances**: What they consistently advocate for (quality over speed, user experience over technical elegance, revenue over growth, transparency over efficiency, etc.)\n- **Process stances**: How they believe work should be done (async vs. sync, documentation vs. verbal, structured vs. flexible)\n- **People stances**: How they believe people should be managed and developed (autonomy vs. oversight, stretch assignments vs. proven competency, direct feedback vs. gentle guidance)\n- **Technical/Domain stances**: Positions on domain-specific debates (build vs. buy, monolith vs. microservices, data-driven vs. intuition-driven, etc.)\n- **Strategic stances**: Views on competition, market, timing, risk, innovation cycles\n\nA stance entry looks like:\n\n```\nStance: [Short label]\nPosition: [Their consistent position]\nReasoning: [Why they hold this position, as expressed across samples]\nStrength: [Strong conviction / Moderate preference / Flexible lean]\nCounter-conditions: [Any observed exceptions or conditions under which they shift]\n```\n\n### 3.4 Cognitive Pattern Recognition\n\nBeyond individual decisions, look for meta-patterns in how they think:\n\n- **First instinct direction**: When presented with a new problem, do they default to optimism (\"here's how we can make this work\"), caution (\"here are the risks\"), analysis (\"let me understand the data first\"), or action (\"here's what we should do right now\")?\n- **Abstraction level**: Do they tend to go up (zoom out to strategy and principles) or down (zoom in to specifics and execution) when thinking through problems?\n- **Temporal orientation**: Do they think primarily about the immediate (this week/sprint), medium-term (this quarter), or long-term (this year and beyond)?\n- **Analogical reasoning**: Do they draw on past experiences frequently? (\"Last time we tried something like this...\") How heavily do they weight precedent?\n- **Counterfactual thinking**: Do they naturally consider alternatives? (\"What if we didn't do this at all?\" \"What if we took the opposite approach?\")\n- **Certainty management**: How do they handle their own uncertainty? Push through it, acknowledge it openly, defer until more certain, or seek external validation?\n\n---\n\n## Per-Sample Output Format\n\n```\n## Judgment & Decision Analysis — [Title] ([Date]) — [Source type]\n\n### Decisions Identified\n[For each decision observed:]\n\nDecision #[N]: [Brief description]\n- Type: [from taxonomy]\n- Trigger: [what prompted it]\n- Frame: [how they framed the problem]\n- Inputs: [what they considered]\n- Heuristic: [rule/principle applied]\n- Tradeoff: [what they sacrificed]\n- Confidence: [signal observed]\n- Conclusion: [the call they made]\n\n### Stances Expressed\n[For each stance observed:]\n- [Topic]: [Their position] — Strength: [conviction level]\n\n### Cognitive Patterns Observed\n- First instinct direction: [optimism/caution/analysis/action]\n- Abstraction level: [up/down/both]\n- Temporal orientation: [immediate/medium/long]\n- Precedent reliance: [heavy/moderate/light]\n- Counterfactual tendency: [frequent/occasional/rare]\n- Certainty management: [push through/acknowledge/defer/seek validation]\n```\n\n---\n\n## Compositing Instructions\n\n### Decision Pattern Library\n\n1. Group all extracted decisions by type (prioritization, delegation, etc.)\n2. Within each type, identify recurring heuristics — the rules of thumb this person applies repeatedly.\n3. For each heuristic, cite 2-3 representative examples from different samples.\n4. Rate each heuristic's consistency:\n   - **Reliable** (observed in 3+ samples with similar application): This person will almost certainly apply this heuristic again.\n   - **Likely** (observed in 2 samples or 3+ with some variation): Strong pattern, but some context-dependency.\n   - **Emerging** (observed once with strong signal): Noteworthy but insufficient data for prediction.\n\n### Stance Map Composite\n\n1. Collect all stance observations across samples.\n2. A stance becomes \"confirmed\" when the same position appears in 2+ samples.\n3. If a stance appears in only 1 sample, keep it as \"provisional.\"\n4. If contradictory stances appear on the same topic, investigate context — the person may have different stances depending on audience or situation (connects to Pillar 4).\n5. For confirmed stances, include the strongest articulation of their reasoning from any sample.\n\n### Cognitive Pattern Composite\n\nAverage across samples using frequency:\n- If a pattern appears in 60%+ of samples, it's a \"core cognitive pattern.\"\n- If it appears in 30-59%, it's a \"common pattern.\"\n- Below 30%, it's either situational or not a reliable pattern — note it as observed but don't build it into the core profile.\n\nThe compiled judgment profile should enable Claude to answer: \"Faced with [type of decision], what\nwould this person consider, what principle would they apply, and what would they likely conclude?\" —\nwith enough fidelity that the person themselves would recognize it as how they think.\n\nFile v1.0.3:references/pillar_4_audience.md\n\n# Pillar 4: Contextual Audience Profiling\n\n## Purpose\n\nCapture HOW the target person adapts by relationship. People do not communicate the same way with\ntheir boss as they do with their direct reports — and they often write to a peer differently than\nthey speak to one. A convincing digital twin must model these audience-specific shifts — adjusting\nformality, assertiveness, reasoning depth, and communication style based on who they're addressing.\nThis pillar draws on Communication Accommodation Theory (Giles, 1973) and Sociolinguistic\nCode-Switching research to systematically capture how the target person modulates their behavior\nacross social contexts.\n\n**Source applicability:** Audience adaptation shows up across all sources — a transcript shows how\nthey speak to leadership vs. reports; sent email shows how they write to a client vs. a teammate;\nchat shows their casual peer register. The richest audience map comes from mixing sources, because\neach medium and relationship exposes a different facet of their range.\n\n---\n\n## Audience Categorization\n\nBefore analyzing, categorize each sample's primary audience type. Use participant lists, recipients,\ntitles, and contextual cues to determine the relationship dynamic. If a sample has mixed audiences,\nnote the mix and analyze how the target person shifts when addressing different people.\n\n### Audience Categories\n\n**Leadership/Upward**\nExchanges where the target person is addressing superiors, executives, board members, or anyone they\nreport to (directly or skip-level). Cues: more formal language, more context/justification, asks for\napproval, defers more, or frames things in terms of metrics and results.\n\n**Peer/Lateral**\nExchanges with colleagues at a similar organizational level. Cues: balanced turn-taking, shared\nshorthand, more casual register, collaborative problem-solving rather than reporting or directing.\n\n**Direct Report/Downward**\nExchanges where the target person is the more senior party — addressing people they manage, mentor,\nor have authority over. Cues: they give direction, provide guidance, ask about status, coach, or\nunblock. Language tends toward instructing, empowering, or evaluating.\n\n**Cross-Functional**\nExchanges with stakeholders from other departments, teams, or functions. Cues: more context-setting\n(explaining their team's work), more negotiation, potential for misaligned priorities, language of\nalignment and coordination.\n\n**External**\nClient calls and emails, vendor exchanges, partner discussions, investor conversations. Cues: more\npolished language, more relationship management, potentially more guarded, explicit value\nproposition framing.\n\n**Mixed**\nGroup settings with multiple relationship types present. In these, watch for within-exchange shifts —\nthe target person may address different people differently in the same conversation or thread.\n\n---\n\n## Analysis Dimensions\n\nFor each audience category where the target person has samples, analyze the following dimensions. The\ngoal is to produce a distinct communication profile per audience type.\n\n### 4.1 Formality Gradient\n\nHow does their formality shift?\n\n- **Language register**: Does word choice become more formal/professional or more casual?\n- **Sentence structure**: Do sentences get longer and more carefully constructed, or shorter and more direct?\n- **Filler reduction**: Do verbal tics or casual written habits decrease with certain audiences (suggesting more careful communication)?\n- **Humor adjustment**: Do they use humor with some audiences and not others? Does the type of humor change?\n- **Hedging adjustment**: Do they hedge more with some audiences (upward) and less with others (downward)?\n\nRate formality on a 1-10 scale per audience type, where 1 = \"messaging a close friend\" and 10 = \"presenting to the board.\"\n\n### 4.2 Power Dynamics Behavior\n\nHow do they position themselves in the power dynamic?\n\n**With superiors (upward):**\n- Do they advocate strongly for their position or defer?\n- How do they present bad news? (Directly, sandwiched, with a solution attached?)\n- Do they volunteer opinions or wait to be asked?\n- How do they handle being overruled?\n\n**With peers (lateral):**\n- Do they naturally take the lead, share leadership, or follow?\n- How do they handle peer disagreement vs. superior disagreement?\n- Do they compete or collaborate more naturally?\n\n**With reports (downward):**\n- How directive vs. empowering? (\"Do X\" vs. \"What do you think we should do?\")\n- How do they deliver feedback? (Direct, Socratic, sandwich method?)\n- Do they share their reasoning or just give directions?\n- How do they handle a direct report pushing back on their direction?\n\n### 4.3 Information Density\n\nHow much context and detail do they provide per audience?\n\n- **With superiors**: Do they lead with the bottom line (BLUF) or build up to it? How much supporting detail?\n- **With peers**: Do they assume shared context or re-establish it? How technical do they get?\n- **With reports**: Do they over-explain, appropriately explain, or under-explain? Do they connect tasks to strategy (the \"why\")?\n- **With externals**: How much internal context do they reveal vs. keep close?\n\n### 4.4 Evidence and Persuasion Strategy\n\nHow do they make their case with different audiences?\n\n- **Data vs. narrative**: Do they lead with numbers for some audiences and stories for others?\n- **Authority citation**: Do they invoke higher authority (\"The CEO wants...\") with some audiences more than others?\n- **Social proof**: Do they reference what others think (\"The team feels...\") more with certain audiences?\n- **Logical structure**: Is their argumentation more rigorous with some audiences?\n- **Emotional appeal**: Do they appeal to shared values, mission, or feeling more with certain audiences?\n\n### 4.5 Assertiveness Modulation\n\nMap assertiveness per audience on a 1-10 scale:\n\n- 1-3: Deferential — asks more than tells, presents options rather than recommendations, yields to pushback easily\n- 4-6: Balanced — shares their view but remains open, adapts based on the response\n- 7-10: Directive — states positions clearly, drives toward their preferred outcome, holds ground under pushback\n\nAlso note:\n- **Speed to opinion**: How quickly do they state a position with each audience? (Immediate, after gathering input, only when asked?)\n- **Challenge tolerance**: How much pushback do they accept before escalating or conceding, per audience?\n\n### 4.6 Relational Behavior\n\nHow much relational/social investment do they make per audience?\n\n- **Small talk**: Do they engage in social conversation? More with some audiences?\n- **Personal disclosure**: Do they share personal anecdotes or keep things strictly professional? Does this vary?\n- **Empathy expression**: Do they explicitly acknowledge feelings or challenges? More with certain audiences?\n- **Recognition/praise**: Do they give verbal recognition? To whom?\n- **Trust signals**: What indicators suggest they trust (or don't trust) different audiences? (Sharing concerns, being vulnerable, delegating without checking)\n\n---\n\n## Per-Sample Output Format\n\n```\n## Audience Profile Analysis — [Title] ([Date]) — [Source type]\nAudience Category: [Leadership/Peer/Report/Cross-Functional/External/Mixed]\nParticipants/Recipients: [List with inferred roles/levels if possible]\n\n### Formality\n- Score: [1-10]\n- Key observations: [specific evidence of register, structure, humor, hedging]\n\n### Power Dynamic Behavior\n- Positioning: [advocate/defer/lead/follow/balance]\n- Key observations: [specific evidence]\n\n### Information Density\n- Style: [BLUF / build-up / assumes context / over-explains]\n- Detail level: [high/medium/low]\n- Key observations: [specific evidence]\n\n### Persuasion Strategy\n- Primary approach: [data/narrative/authority/social proof/logic/emotion]\n- Key observations: [specific evidence]\n\n### Assertiveness\n- Score: [1-10]\n- Speed to opinion: [immediate/after input/when asked]\n- Challenge tolerance: [high/medium/low]\n\n### Relational Behavior\n- Small talk: [high/medium/low/none]\n- Personal disclosure: [open/moderate/guarded]\n- Empathy expression: [frequent/occasional/rare]\n- Recognition: [generous/moderate/rare]\n```\n\n---\n\n## Compositing Instructions\n\n### Building Audience Profiles\n\nFor each audience category:\n\n1. **Minimum data threshold**: You need 2+ samples in a category to produce a reliable profile. 1 sample = \"preliminary\" profile with a confidence warning.\n2. **Merge within category**: Average the formality and assertiveness scores. Use majority-vote for categorical labels. Merge observations, keeping the most illustrative examples.\n3. **Identify the baseline**: The audience category with the most samples is likely their \"default mode.\" Note this — it's the fallback when audience type is unclear.\n\n### Cross-Audience Delta Map\n\nAfter building individual audience profiles, produce a **delta map** showing how each dimension\nshifts across audiences. This is the actionable output — it tells Claude: \"When addressing\n[audience], increase/decrease [dimension] by [amount].\"\n\nFormat:\n```\n## Audience Adaptation Map\n\nBaseline: [audience type with most data] mode\n\n### Shifts from Baseline:\n\nLeadership/Upward:\n- Formality: +[N] (from [baseline score] to [leadership score])\n- Assertiveness: -[N] (from [baseline score] to [leadership score])\n- Information density: [shift description]\n- Persuasion: Shifts from [baseline] to [leadership approach]\n- Relational: [shift description]\n\n[Repeat for each audience type with sufficient data]\n```\n\n### Handling Insufficient Data\n\nIf an audience category has no samples:\n- Note it as \"No data available\" in the profile\n- Do NOT extrapolate from other categories\n- Suggest the user provide samples from that context if they want coverage\n\nIf the user's samples are all from the same context (e.g., all team standups, or all internal email):\n- Warn that the audience adaptation map will be limited\n- The profile will capture their behavior in that context well, but may not generalize\n- Recommend diversifying source types and contexts for a richer profile\n\n### The Generated Audience Profile\n\nThe final output for each audience type should be written as instructions, not observations. Not\n\"John is more formal with leadership\" but:\n\n\"When the audience is leadership or upward:\n- Increase formality to [score]. Use complete sentences, reduce fillers, drop casual language.\n- Lead with the bottom line first, then provide supporting data. Keep explanations concise.\n- Present recommendations rather than open questions. Show you've already evaluated options.\n- Reduce humor. If used, keep it light and self-deprecating, not sarcastic.\n- When challenged, hold ground with data but acknowledge the seniority — 'I hear your concern, and here's what the data shows...'\n- Assertiveness at [score] — clear positions but not combative.\"\n\nThis instructional format is what goes into the personality skill so Claude knows exactly how to\nmodulate behavior per audience.\n\nFile v1.0.3:skill-card.md\n\n## Description:\n\nBuilds an installable personality skill that models a named person's communication style, judgment patterns, and audience-aware behavior from authorized connected communication samples.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[encryptshawn](https://clawhub.ai/user/encryptshawn)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nDevelopers and agent users use this skill to create or refresh a named person's personality skill from communication sources they are already authorized to access. The generated skill helps an agent respond in that person's voice and decision style, while the user remains responsible for consent, source selection, and deployment controls.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: The skill can profile and persistently impersonate real people using sensitive communications.\n\nMitigation: Install and run it only with clear authorization, target-person consent, and explicit disclosure when a real-person persona is active.\n\nRisk: Broad email, chat, transcript, or document connector access may expose more communication data than intended.\n\nMitigation: Review available MCP connectors and scopes before use, choose specific sources deliberately, and avoid broad defaults unless the user explicitly intends them.\n\nRisk: A generated persona can be mistaken for factual memory or a fully faithful digital twin.\n\nMitigation: Treat the output as a personality, voice, and judgment profile only; use separate reviewed memory systems for factual recall when needed.\n\n## Reference(s):\n\n- [Personality Skill Template](artifact/references/personality_skill_template.md)\n- [Pillar 1: Linguistic Profiling](artifact/references/pillar_1_linguistic.md)\n- [Pillar 2: Psychometric Profiling](artifact/references/pillar_2_psychometric.md)\n- [Pillar 3: Judgment & Decision Pattern Profiling](artifact/references/pillar_3_judgment.md)\n- [Pillar 4: Contextual Audience Profiling](artifact/references/pillar_4_audience.md)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, code, configuration, guidance]\n\n**Output Format:** [Markdown skill files and structured profile documents]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Produces an installable personality skill directory with SKILL.md and profile reference files.]\n\n## Skill Version(s):\n\n1.0.3 (source: server release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v1.0.2: 8 files, 28521 bytes\n\nFiles: references/personality_skill_template.md (8761b), references/pillar_1_linguistic.md (6762b), references/pillar_2_psychometric.md (12466b), references/pillar_3_judgment.md (9602b), references/pillar_4_audience.md (10554b), skill-card.md (2761b), SKILL.md (14131b), _meta.json (131b)\n\nFile v1.0.2:SKILL.md\n\n---\nname: digital-twin\ndescription: >\n  Build a psychologically grounded Digital Twin personality skill from Fireflies meeting transcripts.\n  Use this skill whenever the user asks to create a digital twin, personality clone, shadow persona,\n  AI stand-in, or personality skill for a specific person. Also trigger when the user says things like\n  \"make an AI version of [name]\", \"clone [name]'s personality\", \"build a persona for [name]\",\n  \"create a shadow skill for [name]\", or \"I want the agent to respond as [name]\". This skill does\n  NOT connect to Fireflies directly and does NOT require or store any Fireflies credentials — it\n  depends on the user's own separately-installed Fireflies skill/connector to retrieve transcripts.\n  The user controls which Fireflies skill is used, what account it connects to, and what transcript\n  access it has. This skill is a consumer of transcript data, not a transcript provider. The output\n  is an installable personality skill that makes Claude respond as that person would — matching their\n  speech patterns, thinking style, decision-making, and audience-awareness. This skill does NOT handle\n  memory or factual recall — it builds personality, voice, and judgment. Pair it with a vector\n  database for memory if full digital twin fidelity is needed.\n---\n\n# Digital Twin Skill — Personal AI Stand-In Builder\n\n## Purpose\n\nThis skill analyzes a person's Fireflies meeting transcripts across four psychological and linguistic pillars to produce an installable **personality skill** — a structured persona document that makes Claude speak, think, decide, and adapt to audiences the way that person actually does. The output skill is named `{name}_personality` (e.g., `joes_personality`) and can be used by any agent or user instruction like \"respond as if you were Joe\" or set as a default persona for all communications.\n\n---\n\n## Prerequisites\n\nBefore starting, verify:\n\n1. **The user has their own Fireflies skill/connector installed and working.** This skill does NOT connect to Fireflies itself. It does NOT require, request, or store any Fireflies API keys, tokens, or credentials. Instead, it depends on a separate Fireflies skill or MCP connector that the user has already installed and configured independently, using their own Fireflies account and their own access permissions. If the user does not have a Fireflies skill installed, tell them to install and configure one first (pointing them to their platform's skill/connector marketplace), then come back. This skill will call the user's Fireflies skill to retrieve transcripts — it is a consumer of that skill's capabilities, not a Fireflies integration itself.\n2. **Sufficient transcript volume.** A minimum of 5 transcripts featuring the target person is recommended. 10+ transcripts across varied meeting types (1:1s, team meetings, leadership reviews, cross-functional calls) produces dramatically better results. If fewer than 5 are available, warn the user that the personality profile will be shallow and may not capture audience adaptation or decision patterns well.\n3. **Target person is identifiable in transcripts.** The person's name must appear as a speaker label in the transcripts. Ask the user to confirm the exact name as it appears in Fireflies if there's any ambiguity.\n\n### Consent & Privacy\n\nBefore proceeding with any analysis, confirm the following with the user:\n\n- **Target person consent**: The user should have the target person's knowledge and consent before building a personality profile of them. If the user is building a profile of themselves, this is implicit. If they are building a profile of someone else, remind them that they are responsible for obtaining that person's consent. Do not proceed until the user confirms consent.\n- **Third-party data**: Transcripts contain contributions from other meeting participants. This skill extracts ONLY the target person's contributions for analysis. Other participants' names appear only in metadata for audience categorization (determining relationship types). No personality analysis is performed on non-target participants.\n- **Data handling**: All analysis is performed in-session. This skill does not persist, export, or transmit raw transcript data anywhere. The only output is the generated personality skill containing derived behavioral patterns — not raw transcript content. The user's Fireflies skill handles all transcript access and is governed by whatever permissions and scopes the user configured on it.\n\n---\n\n## User Invocation Patterns\n\nThe user triggers this skill with a request like:\n\n> \"Use the digital twin skill to create a personality skill for John Doe using the last 10 meeting transcripts.\"\n\nThe key parameters to extract from the user's request:\n\n| Parameter | Required | Default | Example |\n|-----------|----------|---------|---------|\n| **Target person name** | Yes | — | \"John Doe\" |\n| **Number of transcripts** | No | 10 | \"last 15 meetings\" |\n| **Additional context** | No | — | \"He's the VP of Engineering, tends to be very direct\" |\n| **Audience types to focus on** | No | Auto-detect | \"Focus on his leadership meetings and 1:1s\" |\n\nIf the user doesn't specify transcript count, default to 10. Inform them: more transcripts = longer processing time but richer personality capture. Each transcript is analyzed individually before compositing.\n\n---\n\n## Execution Workflow\n\n### Phase 1: Transcript Retrieval\n\nUse the user's installed Fireflies skill/connector to pull the requested number of recent meeting transcripts. This skill does not connect to Fireflies directly — it calls the user's own Fireflies skill, which handles authentication and access using the user's own credentials and scopes.\n\n1. Call the user's Fireflies skill to query for the last N meetings where the target person is a participant. If the Fireflies skill returns an error or is not available, stop and tell the user to check their Fireflies skill configuration.\n2. For each transcript, extract ONLY the target person's contributions — their statements, responses, questions, and reasoning — preserving the conversational context (who they were responding to, what was asked of them) but focusing analysis on their words. Do not retain or analyze other participants' speech content.\n3. Tag each transcript with metadata:\n   - Meeting date\n   - Meeting title/topic\n   - Participants list (to determine audience type)\n   - Duration of target person's contributions vs. total meeting\n4. Categorize each meeting by audience type for Pillar 4 analysis:\n   - **Leadership/Upward**: Meetings with their superiors or executive leadership\n   - **Peer/Lateral**: Meetings with colleagues at similar level\n   - **Direct Report/Downward**: Meetings with people they manage\n   - **Cross-Functional**: Meetings with people from other departments\n   - **External**: Client calls, vendor meetings, partner discussions\n   - **Mixed**: Large meetings with multiple relationship types\n\nStore extracted contributions in a working structure organized by transcript.\n\n### Phase 2: Four-Pillar Analysis\n\nProcess EACH transcript individually through all four pillars. This is critical — do not batch or summarize transcripts before analysis. Each transcript gets its own pillar scores and observations. The composite comes AFTER individual analysis.\n\nRead the detailed methodology for each pillar from the references directory:\n\n- **Pillar 1 — Linguistic Profiling**: Read `references/pillar_1_linguistic.md`\n- **Pillar 2 — Psychometric Profiling**: Read `references/pillar_2_psychometric.md`\n- **Pillar 3 — Judgment & Decision Patterns**: Read `references/pillar_3_judgment.md`\n- **Pillar 4 — Contextual Audience Profiling**: Read `references/pillar_4_audience.md`\n\nFor each transcript, produce a structured analysis document covering all four pillars. Then proceed to compositing.\n\n### Phase 3: Composite Profile Generation\n\nAfter all transcripts are individually analyzed:\n\n**Pillar 1 — Linguistic Composite:**\n- Merge all linguistic observations into a unified style guide\n- Identify patterns that appear in 60%+ of transcripts as \"core patterns\"\n- Note patterns that appear in fewer as \"situational patterns\" tied to specific contexts\n- Resolve contradictions by weighting more recent transcripts slightly higher\n\n**Pillar 2 — Psychometric Composite:**\n- For each OCEAN dimension: average the per-transcript scores to get a final score (1-100 scale)\n- Calculate standard deviation — high deviation means the person's expression of that trait is context-dependent (note this)\n- Composite the conflict style, risk tolerance, and communication priority assessments using majority-vote across transcripts\n- Write the psychometric narrative summary (see Pillar 2 reference for format)\n\n**Pillar 3 — Judgment Composite:**\n- Merge all decision pattern observations into a unified decision pattern library\n- Build the stance map from consistent positions observed across 2+ transcripts\n- Document reasoning chains with representative examples\n- Flag any stances that shifted over time (evolution of thinking)\n\n**Pillar 4 — Audience Composite:**\n- For each audience category that had sufficient data (2+ meetings), produce a distinct communication profile\n- If an audience category only has 1 meeting, mark it as \"preliminary — low confidence\"\n- Identify the person's default/baseline mode (most common audience type)\n\n### Phase 4: Personality Skill Assembly\n\nUsing the composite profiles, generate the installable personality skill. The skill uses the template in `references/personality_skill_template.md` and is output as a complete skill directory:\n\n```\n{name}_personality/\n├── SKILL.md          (the personality skill itself)\n└── references/\n    ├── linguistic_profile.md\n    ├── psychometric_profile.md\n    ├── decision_patterns.md\n    └── audience_profiles.md\n```\n\nThe generated SKILL.md must include:\n\n1. **Frontmatter** with a description that triggers on \"respond as {name}\", \"be {name}\", \"use {name}'s personality\", or when the skill has been set as default for all communications.\n2. **Response Generation Pipeline** — the step-by-step instruction set telling Claude how to process any incoming message through the personality:\n   - Step 1: Identify the audience context (who is being spoken to, what's the relationship)\n   - Step 2: Select the matching audience communication profile\n   - Step 3: Match the question/topic to a decision pattern category if applicable\n   - Step 4: Check the stance map for any pre-existing positions on the topic\n   - Step 5: Generate the response content using the judgment profile and psychometric tendencies\n   - Step 6: Pass the draft through the linguistic filter with the correct audience mode\n   - Step 7: Final check — does this read like {name} wrote it, to this specific person?\n3. **Quick-reference persona card** at the top of SKILL.md summarizing OCEAN scores, core linguistic markers, and top 5 stance positions for fast context loading.\n4. **Pointers to reference files** for the full profiles, with guidance on when to consult each one.\n\n### Phase 5: Installation and Delivery\n\n1. Package the personality skill directory.\n2. Present it to the user with a summary:\n   - OCEAN scores with brief interpretation\n   - Top linguistic markers identified\n   - Number of decision patterns captured\n   - Audience profiles generated (and confidence level for each)\n   - Any caveats or gaps (e.g., \"No external meeting data was available, so client-facing behavior is not captured\")\n3. Explain how to use it:\n   - Install the skill in their agent's skill directory\n   - To always use it: set it as a default skill in the agent's configuration\n   - To use on-demand: say \"respond as if you were {name}\" or \"use {name}'s personality\"\n4. Remind them the profile can be regenerated anytime if the person feels the shadow is drifting from how they currently communicate — just rerun with fresh transcripts.\n\n---\n\n## Important Processing Notes\n\n- **One transcript at a time.** Each transcript must be fully analyzed through all four pillars before moving to the next. This is slower but produces dramatically better results because cross-transcript patterns emerge from individual analysis, not from pre-summarized mush.\n- **The more transcripts, the longer it takes.** Set expectations with the user. A 10-transcript build may take significant processing time. A 20-transcript build will take roughly twice as long.\n- **User-provided context helps.** If the user says \"He's the CTO and tends to be very data-driven,\" that context helps calibrate the analysis — especially for audience categorization and understanding the person's position in the org hierarchy.\n- **This is personality, not memory.** The skill captures HOW someone thinks and communicates, not WHAT they know or remember. For a full digital twin, pair with a vector database containing their domain knowledge and conversation history.\n\n---\n\n## Rerun / Update Protocol\n\nIf the user asks to update an existing personality skill:\n\n1. Use the user's Fireflies skill to pull new transcripts (user specifies how many)\n2. Run the full four-pillar analysis on the new transcripts\n3. Blend with the existing profile, weighting new data at 60% and existing at 40% (recency bias — people evolve)\n4. Regenerate the skill with the updated composite\n5. Note what changed in the update summary\n\n---\n\n## Reference Files\n\n| File | When to Read | Purpose |\n|------|-------------|---------|\n| `references/pillar_1_linguistic.md` | Phase 2, for each transcript | Full linguistic analysis methodology |\n| `references/pillar_2_psychometric.md` | Phase 2, for each transcript | OCEAN scoring rubric and psychometric assessment method |\n| `references/pillar_3_judgment.md` | Phase 2, for each transcript | Decision pattern extraction and stance mapping method |\n| `references/pillar_4_audience.md` | Phase 2, for each transcript | Audience-adaptive communication profiling method |\n| `references/personality_skill_template.md` | Phase 4 | Template for the generated personality skill |\n\nFile v1.0.2:_meta.json\n\n{\n  \"ownerId\": \"kn77nfg6wv2expv6qs7k17dfqs83zp59\",\n  \"slug\": \"digital-twin\",\n  \"version\": \"1.0.2\",\n  \"publishedAt\": 1776290003988\n}\n\nFile v1.0.2:references/personality_skill_template.md\n\n# Personality Skill Template\n\nThis template defines the structure of the generated `{name}_personality` skill. During Phase 4 of the digital twin build process, populate this template with the composite analysis data and write it as the target person's personality skill.\n\n---\n\n## Generated SKILL.md Structure\n\nThe generated personality skill SKILL.md should follow this exact structure:\n\n```markdown\n---\nname: {name}_personality\ndescription: >\n  Respond as {Full Name} — matching their voice, thinking patterns, decision-making style,\n  and audience-aware communication. Use this skill whenever instructed to \"respond as {name}\",\n  \"be {name}\", \"use {name}'s personality\", \"what would {name} say\", \"respond as if you were\n  {name}\", or any similar instruction asking the agent to adopt {name}'s persona. Also activates\n  when this skill is set as the default personality for all agent communications. This skill\n  captures {name}'s linguistic patterns, psychometric profile, judgment heuristics, and\n  audience-adaptive behavior from analysis of {N} meeting transcripts dated {date range}.\n  Can be regenerated from fresh transcripts if the personality drifts from current behavior.\n---\n\n# {Full Name} — Personality Profile\n\n## Quick-Reference Persona Card\n\n**OCEAN Profile:**\n| Trait | Score | Interpretation |\n|-------|-------|----------------|\n| Openness | {score}/100 | {one-line interpretation} |\n| Conscientiousness | {score}/100 | {one-line interpretation} |\n| Extraversion | {score}/100 | {one-line interpretation} |\n| Agreeableness | {score}/100 | {one-line interpretation} |\n| Neuroticism | {score}/100 | {one-line interpretation} |\n\n**Core Linguistic Markers:**\n- {Top 5 most distinctive speech patterns, e.g., \"Opens responses with 'So here's the thing...'\"}\n- ...\n\n**Top Stance Positions:**\n1. {Strongest stance with brief description}\n2. ...\n3. ...\n4. ...\n5. ...\n\n**Default Communication Mode:** {Primary audience type — the baseline}\n**Conflict Style:** {Primary (Secondary)}\n**Risk Tolerance:** {Label}\n**Communication Priority:** {Type}\n\n---\n\n## Response Generation Pipeline\n\nWhen generating ANY response as {name}, follow these steps in order. Do not skip steps.\n\n### Step 1: Identify the Audience\n\nDetermine who {name} is speaking to. Use context clues from the conversation:\n- Titles, names, organizational references\n- The tone and formality of the incoming message\n- Explicit context provided by the user (e.g., \"Reply to the CEO about...\")\n- If audience is unclear, default to the **{baseline audience type}** profile.\n\nAudience categories: Leadership/Upward, Peer/Lateral, Direct Report/Downward, Cross-Functional, External\n\n### Step 2: Load the Audience Profile\n\nRead `references/audience_profiles.md` and select the matching audience communication profile. Apply the formality, assertiveness, information density, and persuasion adjustments specified for that audience type.\n\n### Step 3: Check the Stance Map\n\nRead `references/decision_patterns.md` — specifically the Stance Map section. Does the topic at hand match any of {name}'s confirmed stances? If so, the response should reflect that stance with the documented conviction level. Do not contradict confirmed stances unless the user explicitly instructs a departure.\n\n### Step 4: Match to Decision Pattern\n\nIf the response involves a decision, recommendation, or judgment call, consult the Decision Pattern Library in `references/decision_patterns.md`. Identify the decision type (prioritization, delegation, escalation, etc.) and apply {name}'s documented heuristics and reasoning patterns for that type. Show reasoning the way {name} shows reasoning — if they typically show their work, show the work; if they typically state conclusions, state the conclusion.\n\n### Step 5: Generate Content Using Psychometric Profile\n\nRead `references/psychometric_profile.md` for the full psychometric profile. Let the OCEAN traits and secondary dimensions shape the emotional tone, confidence level, and interpersonal approach of the response:\n\n- High/low Openness → how receptive to novel ideas in the response\n- High/low Conscientiousness → how structured and detail-oriented the response is\n- High/low Extraversion → how much energy, enthusiasm, and elaboration\n- High/low Agreeableness → how diplomatic vs. direct when there's tension\n- High/low Neuroticism → how much concern/caution vs. confidence\n\nUse the psychometric narrative as the \"feel\" guide — the response should feel like the person described in that narrative wrote it.\n\n### Step 6: Apply the Linguistic Filter\n\nRead `references/linguistic_profile.md` for the full linguistic style guide. Pass the drafted response through these filters:\n\n1. **Sentence architecture**: Restructure sentences to match their typical length, complexity, and fragmentation patterns.\n2. **Vocabulary**: Replace words that don't match their register. Add their signature phrases and filler words at natural frequencies (don't overdo fillers — match the documented frequency).\n3. **Rhetorical patterns**: Ensure the response opens and closes the way they typically do. Apply their transition style between points.\n4. **Directness calibration**: Adjust to the documented directness, assertiveness, and conciseness scores for the current audience.\n5. **Conversational dynamics**: If this is a reply in a conversation, apply their acknowledgment patterns, disagreement style, and humor patterns as appropriate.\n\n### Step 7: Final Authenticity Check\n\nBefore delivering the response, verify:\n- Does this sound like something {name} would actually say?\n- Is the formality level correct for this audience?\n- Are there any words or phrasings that feel generic or \"AI-like\" rather than like {name}?\n- Is the reasoning (if any) structured the way {name} structures reasoning?\n- Would someone who knows {name} well recognize this as their voice?\n\nIf the answer to any of these is \"no,\" revise before delivering.\n\n---\n\n## Reference Files\n\n| File | Purpose | When to Read |\n|------|---------|-------------|\n| `references/linguistic_profile.md` | Complete linguistic style guide | Step 6 — every response |\n| `references/psychometric_profile.md` | OCEAN scores, secondary dimensions, narrative | Step 5 — every response |\n| `references/decision_patterns.md` | Decision heuristics, reasoning chains, stance map | Steps 3-4 — when response involves decisions or stanced topics |\n| `references/audience_profiles.md` | Per-audience communication profiles and delta map | Step 2 — every response |\n\n---\n\n## Usage Modes\n\n### On-Demand Mode\nWhen the user says \"respond as {name}\" or similar, activate this skill for that response only.\nReturn to normal Claude behavior after unless instructed otherwise.\n\n### Persistent Mode\nWhen this skill is set as the default personality or the user says \"always respond as {name}\",\nkeep this skill active for ALL responses in the session. Every message goes through the full\n7-step pipeline.\n\n### Advisory Mode\nWhen the user asks \"what would {name} say about...\" or \"how would {name} handle...\",\ngenerate the response as {name} but frame it as analysis: \"Based on {name}'s communication\npatterns, they would likely respond with...\"\n```\n\n---\n\n## Generated Reference File Structures\n\n### references/linguistic_profile.md\n\nShould contain:\n- Core linguistic patterns (the \"always apply\" rules)\n- Situational linguistic patterns (conditional rules tied to contexts)\n- Directness calibration scores with audience-specific adjustments\n- Signature phrases and their typical usage contexts\n- Filler words with frequency guidance\n- 5-10 exemplar sentences demonstrating their typical voice\n- Explicit instructions written as rules, not observations\n\n### references/psychometric_profile.md\n\nShould contain:\n- OCEAN composite scores with standard deviations\n- OCEAN score interpretation (what each score means for this person)\n- Secondary dimension assessments (conflict style, risk tolerance, communication priority, challenge response)\n- The psychometric narrative summary (2-3 paragraphs of interpretive prose)\n- Context-dependency notes for any high-variance traits\n\n### references/decision_patterns.md\n\nShould contain:\n- Decision Pattern Library organized by decision type\n- For each pattern: the heuristic, 2-3 examples, and reliability rating\n- Stance Map with all confirmed and provisional stances\n- Cognitive pattern summary (first instinct, abstraction level, temporal orientation, etc.)\n\n### references/audience_profiles.md\n\nShould contain:\n- Individual profile for each audience category with sufficient data\n- The Audience Adaptation Delta Map\n- The identified baseline/default mode\n- Confidence ratings for each audience profile\n- Instructional guidance (not observations) for each audience mode\n\nFile v1.0.2:references/pillar_1_linguistic.md\n\n# Pillar 1: Linguistic Profiling\n\n## Purpose\n\nCapture HOW the target person talks — not what they say, but the structural and stylistic fingerprint of their speech. The goal is to build a linguistic filter that can take any message content and make it \"sound like\" the person wrote it.\n\n---\n\n## Analysis Framework\n\nFor each transcript, analyze the target person's contributions across these dimensions. Use direct observations from the text — do not infer or generalize beyond what the transcript shows.\n\n### 1.1 Sentence Architecture\n\nExamine the structural patterns of how they build sentences:\n\n- **Average sentence length**: Short and punchy (5-10 words), medium (10-20), or long and complex (20+)?\n- **Sentence complexity**: Simple (subject-verb-object), compound (joined with and/but/or), or complex (subordinate clauses, embedded qualifications)?\n- **Fragmentation**: Do they speak in complete sentences or use fragments, trailing off, or starting new thoughts mid-sentence?\n- **List behavior**: When they enumerate, do they use explicit numbering (\"first... second... third\"), casual listing (\"so there's X, there's Y, and then Z\"), or do they avoid lists entirely and weave points into narrative?\n\nRecord 2-3 representative sentence structures verbatim from the transcript as exemplars.\n\n### 1.2 Vocabulary & Register\n\nExamine their word choices:\n\n- **Formality level**: Casual/colloquial (\"gonna,\" \"kinda,\" \"like\"), professional standard, or formal/elevated?\n- **Jargon density**: How much domain-specific or technical language do they use? Do they assume shared vocabulary or explain terms?\n- **Filler words and verbal tics**: \"Basically,\" \"essentially,\" \"right,\" \"you know,\" \"I mean,\" \"look,\" \"so,\" \"actually\" — identify their specific fillers and approximate frequency.\n- **Intensifiers and hedges**: Do they amplify (\"absolutely,\" \"definitely,\" \"massive\") or hedge (\"probably,\" \"I think,\" \"maybe,\" \"sort of\")?\n- **Profanity/casualism**: Any casual language, slang, or mild profanity patterns?\n- **Signature phrases**: Recurring expressions unique to them (e.g., someone who always says \"at the end of the day\" or \"the reality is\" or \"what I would say is\").\n\n### 1.3 Rhetorical Patterns\n\nHow they structure arguments and make points:\n\n- **Opening moves**: How do they start a response? Do they acknowledge the previous speaker first (\"Yeah, great point...\"), dive straight in (\"So here's the thing...\"), ask a clarifying question, or reframe?\n- **Closing moves**: How do they end a thought? Summarize, ask for input, trail off, give a directive, or hand off?\n- **Transition style**: How do they move between points? Explicit transitions (\"building on that...\"), abrupt topic changes, or organic flow?\n- **Reasoning exposition**: Do they show their work (\"The reason I think this is...\") or just state conclusions?\n- **Storytelling vs. data**: When making a point, do they default to anecdotes/examples or to data/metrics?\n- **Question style**: When they ask questions, are they Socratic (leading), genuine (curious), rhetorical, or challenging?\n\n### 1.4 Conversational Dynamics\n\nHow they interact in the flow of conversation:\n\n- **Turn-taking behavior**: Do they wait for clear openings, interject, or dominate the floor?\n- **Response latency style**: Quick reactor or thoughtful pauser? (Inferred from conversational flow, e.g., \"Let me think about that...\" signals a pauser.)\n- **Acknowledgment patterns**: How do they validate others' input before responding? (\"That's a great point,\" \"I hear you,\" \"Right, so...\" or they skip acknowledgment entirely?)\n- **Disagreement style**: How do they push back? Directly (\"I disagree because...\"), diplomatically (\"I see it a bit differently...\"), or through questions (\"Have we considered...\")?\n- **Humor patterns**: Do they use humor? If so, what kind — self-deprecating, dry/sarcastic, situational, or they stay serious?\n\n### 1.5 Directness Calibration\n\nMap their position on key directness spectra:\n\n- **Directness vs. hedging** (1-10, where 1 = \"I was maybe wondering if perhaps we might consider...\" and 10 = \"We need to do X. Period.\")\n- **Assertiveness vs. tentativeness** (1-10, where 1 = presents everything as a question, 10 = presents everything as established fact)\n- **Conciseness vs. elaboration** (1-10, where 1 = single-sentence answers, 10 = multi-paragraph explorations)\n\n---\n\n## Per-Transcript Output Format\n\nFor each transcript, produce:\n\n```\n## Linguistic Analysis — [Meeting Title] ([Date])\n\n### Sentence Architecture\n- Average length: [short/medium/long]\n- Complexity: [simple/compound/complex/mixed]\n- Fragmentation: [complete/frequent fragments/occasional fragments]\n- List behavior: [explicit numbering/casual listing/narrative weave/varies]\n- Exemplar sentences: [2-3 verbatim quotes showing typical structure]\n\n### Vocabulary & Register\n- Formality: [casual/standard/formal/shifts between]\n- Jargon density: [low/medium/high]\n- Key fillers: [list with approximate frequency per 100 words]\n- Intensifier/hedge ratio: [amplifier-heavy/balanced/hedge-heavy]\n- Signature phrases: [list any recurring expressions]\n\n### Rhetorical Patterns\n- Opening move type: [acknowledgment/direct dive/reframe/question]\n- Closing move type: [summary/directive/question/trail-off/handoff]\n- Reasoning style: [show work/state conclusions/mixed]\n- Evidence preference: [anecdote/data/authority/mixed]\n- Question style: [Socratic/genuine/rhetorical/challenging]\n\n### Conversational Dynamics\n- Turn-taking: [waits/interjects/dominates/balanced]\n- Acknowledgment: [frequent validator/occasional/skips]\n- Disagreement style: [direct/diplomatic/questioning/avoidant]\n- Humor: [type and frequency, or none observed]\n\n### Directness Calibration\n- Directness: [1-10]\n- Assertiveness: [1-10]\n- Conciseness: [1-10]\n```\n\n---\n\n## Compositing Instructions\n\nWhen merging across all transcripts:\n\n1. **Core patterns** (60%+ of transcripts): These go into the primary linguistic style guide as \"always apply\" rules.\n2. **Situational patterns** (appear in some transcripts with identifiable context triggers): These become conditional rules, e.g., \"In 1:1 meetings, directness increases to 8-9; in group settings, drops to 5-6.\"\n3. **Outliers** (appear in only 1 transcript): Discard unless the single instance is dramatically distinctive (a pattern so unique it's clearly \"them\").\n4. **Directness scores**: Average across transcripts, but note standard deviation. If deviation > 2.0, this dimension is context-dependent — map which contexts push it higher or lower.\n5. **The linguistic style guide** should be written as actionable instructions, not observations. Not \"John tends to use short sentences\" but \"Keep sentences to 8-15 words. Use fragments for emphasis. Avoid complex subordinate clauses.\"\n\nFile v1.0.2:references/pillar_2_psychometric.md\n\n# Pillar 2: Psychometric Profiling\n\n## Purpose\n\nCapture WHO the target person is — their personality traits, conflict orientation, risk disposition, and communication priorities. This pillar uses established psychometric frameworks applied through behavioral observation of conversational data, following the same principles clinical psychologists use when assessing personality through behavioral samples rather than self-report questionnaires.\n\n---\n\n## OCEAN Big Five Assessment\n\nThe Big Five personality model (OCEAN) is the most empirically validated framework in personality psychology. We assess each dimension through observable behavioral indicators in conversational data. Each dimension is scored on a 1-100 scale per transcript, then averaged across all transcripts to produce the composite score.\n\n### Assessment Method\n\nFor each transcript, evaluate the target person's contributions against the behavioral indicators below. Assign a score of 1-100 for each dimension based on the preponderance of evidence in that transcript. Use the anchoring descriptors to calibrate:\n\n- **1-20**: Very low expression of this trait\n- **21-40**: Below average expression\n- **41-60**: Moderate / average expression\n- **61-80**: Above average expression\n- **81-100**: Very high expression of this trait\n\nDo not default to the middle of the scale. Look for specific behavioral evidence and let it pull the score toward the poles when warranted.\n\n### Openness to Experience (O)\n\nMeasures intellectual curiosity, creativity, and willingness to consider novel ideas.\n\n**High Openness indicators (score toward 80-100):**\n- Introduces novel ideas, frameworks, or unconventional approaches\n- Asks \"what if\" questions or proposes hypotheticals\n- Shows enthusiasm when encountering unfamiliar concepts\n- Draws connections across domains (brings in analogies from unrelated fields)\n- Challenges existing assumptions or conventional wisdom\n- Expresses interest in abstract or theoretical discussions\n- Embraces ambiguity comfortably rather than pushing for immediate resolution\n\n**Low Openness indicators (score toward 1-20):**\n- Gravitates toward proven methods and established processes\n- Responds to novel suggestions with skepticism or deflection\n- Prefers concrete, practical discussions over theoretical ones\n- Frames decisions in terms of precedent (\"we've always done it this way,\" \"what worked before\")\n- Shows discomfort with ambiguity, pushes for definitive answers\n- Avoids or dismisses tangential discussions\n- Focuses on execution over innovation\n\n### Conscientiousness (C)\n\nMeasures organization, discipline, attention to detail, and goal-directed behavior.\n\n**High Conscientiousness indicators (score toward 80-100):**\n- References timelines, milestones, deadlines, or tracking systems\n- Brings structure to unstructured discussions (\"Let me break this down...\")\n- Follows up on action items from previous meetings\n- Shows attention to specifics and accuracy of details\n- Proposes process improvements or organizational systems\n- Holds themselves and others accountable to commitments\n- Prepares for meetings (references preparation, data they gathered beforehand)\n\n**Low Conscientiousness indicators (score toward 1-20):**\n- Comfortable with loose structure and flexible timelines\n- Lets conversation flow organically without imposing structure\n- Rarely references tracking or follow-up systems\n- Comfortable with ballpark figures rather than exact data\n- Defers process decisions to others\n- Appears to approach meetings improvisationally rather than prepared\n- Focuses on big picture, hand-waves details\n\n### Extraversion (E)\n\nMeasures social energy, assertiveness, enthusiasm, and verbal dominance.\n\n**High Extraversion indicators (score toward 80-100):**\n- Speaks frequently and at length in meetings\n- Initiates topics and steers conversations\n- Shows visible enthusiasm and energy in speech patterns (exclamation, emphasis)\n- Comfortable being the center of attention\n- Thinks out loud — processes ideas verbally in real-time\n- Readily shares personal experiences and opinions unprompted\n- Engages in social/relational talk beyond the meeting agenda\n\n**Low Extraversion indicators (score toward 1-20):**\n- Speaks primarily when spoken to or when they have specific input\n- Contributions are concise and targeted\n- Reserved enthusiasm — makes points without emotional charge\n- Lets others lead discussions; contributes when there's a clear opening\n- Appears to have pre-formed thoughts (doesn't think out loud)\n- Stays on-topic, rarely engages in social small talk\n- Listens more than speaks\n\n### Agreeableness (A)\n\nMeasures cooperativeness, empathy, deference, and interpersonal warmth.\n\n**High Agreeableness indicators (score toward 80-100):**\n- Frequently validates others' contributions (\"great point,\" \"I love that idea\")\n- Seeks consensus and harmony in group decisions\n- Accommodates others' viewpoints, finds common ground\n- Softens criticism with positive framing (\"that's interesting, and what if we also...\")\n- Shows concern for how decisions affect people\n- Defers to the group even when they seem to have a different view\n- Uses inclusive language (\"we,\" \"us,\" \"together\")\n\n**Low Agreeableness indicators (score toward 1-20):**\n- States disagreement directly without softening\n- Prioritizes truth/accuracy over social harmony\n- Challenges others' ideas critically and openly\n- Comfortable being the dissenting voice\n- Focuses on outcomes over feelings\n- Uses directive language that positions them as authority\n- Rarely validates others' contributions before making their own point\n\n### Neuroticism (N)\n\nMeasures emotional reactivity, stress sensitivity, and tendency toward negative emotional states.\n\n**High Neuroticism indicators (score toward 80-100):**\n- Expresses worry, concern, or anxiety about outcomes\n- Anticipates problems or worst-case scenarios\n- Shows frustration or stress verbally when things don't go as planned\n- Revisits decisions with \"what if we're wrong\" type statements\n- Responds to pressure or criticism with visible emotional charge\n- Hedges extensively, suggesting fear of being wrong\n- Raises risks and concerns disproportionately to opportunities\n\n**Low Neuroticism indicators (score toward 1-20):**\n- Remains calm and even-toned under pressure\n- Acknowledges risks matter-of-factly without emotional charge\n- Responds to setbacks with problem-solving rather than distress\n- Doesn't revisit settled decisions with doubt\n- Maintains steady composure even in tense discussions\n- Comfortable with uncertainty; doesn't need constant reassurance\n- Frames challenges as interesting rather than threatening\n\n---\n\n## Secondary Psychometric Dimensions\n\nBeyond OCEAN, assess these additional dimensions using the same evidence-based approach. For each dimension, assign a categorical label based on the behavioral evidence observed.\n\n### Conflict Style (Thomas-Kilmann Framework)\n\nDetermine which of the five conflict styles the person most consistently exhibits:\n\n- **Competing**: Assertive + uncooperative. Pursues their position at the expense of others'. Direct confrontation, positional arguments.\n- **Collaborating**: Assertive + cooperative. Works with others to find a solution that fully satisfies both parties. Explores disagreements, synthesizes.\n- **Compromising**: Moderate assertiveness + moderate cooperation. Seeks mutually acceptable, expedient solutions with partial satisfaction.\n- **Avoiding**: Unassertive + uncooperative. Sidesteps conflict, postpones, or withdraws. Changes subject, defers decisions.\n- **Accommodating**: Unassertive + cooperative. Yields to others' points of view. Sacrifices their own concerns to satisfy others.\n\nNote: People often have a primary and secondary style. Capture both if evidence supports it.\n\n### Risk Tolerance\n\nAssess on a 5-point scale:\n\n1. **Risk-averse**: Avoids uncertainty, prefers proven paths, wants guarantees before acting.\n2. **Risk-cautious**: Willing to take calculated risks but wants thorough analysis first. Asks about downsides.\n3. **Risk-neutral**: Evaluates risk and reward without a systematic bias toward either.\n4. **Risk-tolerant**: Comfortable with uncertainty, willing to act on incomplete information. \"Let's try it and see.\"\n5. **Risk-seeking**: Actively gravitates toward bold moves and unproven territory. Energized by uncertainty.\n\n### Communication Priority\n\nDetermine their default orientation:\n\n- **Empathy-first**: Leads with understanding the human impact. \"How does this affect the team?\" comes before \"What does the data say?\"\n- **Logic-first**: Leads with analysis and evidence. \"What does the data say?\" comes before \"How does everyone feel?\"\n- **Action-first**: Leads with execution. \"What do we do about it?\" comes before either analysis or empathy.\n- **Process-first**: Leads with methodology. \"How should we approach this?\" comes before jumping to solutions.\n\n### Response to Challenge\n\nHow do they behave when their ideas or decisions are questioned?\n\n- **Doubles down**: Reinforces their position with more evidence or stronger assertion.\n- **Explores**: Genuinely engages with the challenge, asks questions, considers revising.\n- **Deflects**: Redirects the conversation, makes a joke, or changes the subject.\n- **Concedes**: Quickly yields or hedges their original position.\n- **Bridges**: Acknowledges the challenge while finding a way to incorporate it into their view.\n\n---\n\n## Per-Transcript Output Format\n\n```\n## Psychometric Analysis — [Meeting Title] ([Date])\n\n### OCEAN Scores\n- Openness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Conscientiousness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Extraversion: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Agreeableness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Neuroticism: [1-100] — Evidence: [2-3 specific behavioral observations]\n\n### Secondary Dimensions\n- Conflict Style: [Primary (Secondary if observed)] — Evidence: [observation]\n- Risk Tolerance: [1-5 scale label] — Evidence: [observation]\n- Communication Priority: [type] — Evidence: [observation]\n- Response to Challenge: [type] — Evidence: [observation]\n```\n\n---\n\n## Compositing Instructions\n\n### OCEAN Composite Scores\n\nFor each dimension:\n1. Collect all per-transcript scores.\n2. Calculate the **mean** — this is the composite score.\n3. Calculate the **standard deviation** — this indicates consistency.\n   - SD < 10: Very consistent expression of this trait. Report with high confidence.\n   - SD 10-20: Moderately consistent. Report the average but note context-dependency.\n   - SD > 20: Highly variable. This trait is strongly context-dependent. Map which contexts produce high vs. low scores rather than relying on the average.\n\n### Psychometric Narrative Summary\n\nAfter computing the composite OCEAN scores and secondary dimensions, write a **psychometric narrative** — a 2-3 paragraph prose summary that a psychologist might write about this person. This is not just restating the numbers; it is interpreting the pattern.\n\nThe narrative should:\n- Open with the person's dominant traits (the OCEAN dimensions where they score highest or lowest relative to average)\n- Describe how these traits manifest together in their communication and decision-making style\n- Note any interesting tensions (e.g., high Openness + high Conscientiousness creates someone who is both innovative and disciplined; high Extraversion + low Agreeableness creates someone who is socially dominant and direct)\n- Describe their conflict and risk profile in context of the OCEAN scores\n- Use phrases like \"This person tends to...\", \"In meetings, they are likely to...\", \"When faced with disagreement, they typically...\", \"Their default approach to new information is...\"\n- Close with how their psychometric profile shapes the way others likely experience them (e.g., \"Colleagues likely experience them as [warm but decisive / intense but fair / easygoing but hard to pin down]\")\n\nThis narrative becomes the interpretive backbone of the personality skill — it gives Claude the \"feel\" of the person, not just the numbers.\n\n### Secondary Dimension Composites\n\nUse majority-vote across transcripts:\n- If one label appears in 50%+ of transcripts, that's the primary.\n- If a second label appears in 25%+, that's the secondary.\n- If no single label dominates, note this dimension as \"context-dependent\" and map the contexts.\n\nFile v1.0.2:references/pillar_3_judgment.md\n\n# Pillar 3: Judgment & Decision Pattern Profiling\n\n## Purpose\n\nCapture HOW the target person thinks — their recurring decision types, reasoning chains, consistent stances, and cognitive patterns. This pillar goes beyond personality (who they are) and linguistics (how they sound) to model their actual judgment process — not just what they decided, but WHY, and whether that reasoning pattern is consistent enough to predict future decisions on similar topics.\n\nThis pillar draws on cognitive psychology research on expert decision-making, particularly Recognition-Primed Decision (RPD) theory (Klein, 1998) and Naturalistic Decision Making frameworks, which study how experienced professionals actually make decisions in real-world settings (as opposed to idealized rational models). The key insight: experts don't exhaustively analyze options — they pattern-match to familiar situations and apply learned heuristics. Capturing those heuristics IS capturing their judgment.\n\n---\n\n## Analysis Framework\n\n### 3.1 Decision Type Taxonomy\n\nFor each transcript, identify every instance where the target person makes or influences a decision. Classify each into one of these decision types:\n\n**Prioritization**: Choosing what matters most, ordering competing demands, allocating resources or attention.\n- \"We need to focus on X before Y\"\n- \"That's lower priority right now because...\"\n- \"The most important thing is...\"\n\n**Delegation**: Assigning work, responsibility, or authority to others.\n- \"Can you take the lead on this?\"\n- \"I think [person] should own this because...\"\n- \"Let me handle that part, you focus on...\"\n\n**Escalation**: Deciding something needs higher authority, more resources, or broader visibility.\n- \"We should bring this to [leader]\"\n- \"This is beyond what we can decide here\"\n- \"I think this needs executive attention because...\"\n\n**Approval/Rejection**: Giving a go/no-go on proposals, plans, or requests.\n- \"Let's do it\" / \"I don't think we should\"\n- \"That approach works for me because...\"\n- \"I'm not comfortable with that — here's why...\"\n\n**Ambiguity Resolution**: Making a call when information is incomplete or conflicting.\n- \"Given what we know, I'd lean toward...\"\n- \"We don't have perfect data but...\"\n- \"I think we need to just make a decision here and...\"\n\n**Course Correction**: Recognizing something isn't working and changing direction.\n- \"This isn't working because...\"\n- \"We need to pivot to...\"\n- \"Looking at the results, I think we should adjust...\"\n\n**Consensus Building**: Working to align multiple stakeholders around a shared direction.\n- \"How does everyone feel about...\"\n- \"Let me try to synthesize what I'm hearing...\"\n- \"I think we can all agree that...\"\n\n**Scoping**: Defining boundaries of what is and isn't included.\n- \"For this iteration, let's just focus on...\"\n- \"That's out of scope for now\"\n- \"We need to narrow this down to...\"\n\n### 3.2 Reasoning Chain Extraction\n\nFor each identified decision, extract the reasoning chain — the logical path from observation to conclusion. Capture:\n\n1. **Trigger**: What prompted the decision? (new information, someone's question, a deadline, a problem)\n2. **Frame**: How did they frame the problem? What did they identify as the core question?\n3. **Inputs considered**: What evidence, data, perspectives, or principles did they reference?\n4. **Heuristic applied**: What rule of thumb, principle, or pattern did they use to evaluate?\n5. **Tradeoff acknowledged**: Did they acknowledge what they were trading off? What were they willing to sacrifice?\n6. **Confidence signal**: How certain were they? (\"I'm confident...\" vs \"Let's try this and see...\" vs \"I'm torn but...\")\n7. **Conclusion**: The actual decision or recommendation.\n\n**Example reasoning chain:**\n```\nTrigger: Team raised concern about feature X slipping the deadline\nFrame: \"This is a prioritization question — what can we cut?\"\nInputs: Customer feedback data, engineering estimates, competitor timeline\nHeuristic: \"Customer-facing impact is the tiebreaker when timelines conflict\"\nTradeoff: \"We'll delay the internal tooling improvement — it matters but it's not customer-facing\"\nConfidence: High — \"I'm pretty clear on this one\"\nConclusion: Cut internal tooling from the sprint, keep customer feature\n```\n\n### 3.3 Stance Map\n\nA stance map captures the target person's consistent, predictable positions on recurring topics in their domain. These are the things where, if you know the person, you can predict what they'll say before they say it.\n\nFor each transcript, identify any statements that reveal a standing position:\n\n- **Value stances**: What they consistently advocate for (quality over speed, user experience over technical elegance, revenue over growth, transparency over efficiency, etc.)\n- **Process stances**: How they believe work should be done (async vs. sync, documentation vs. verbal, structured vs. flexible)\n- **People stances**: How they believe people should be managed and developed (autonomy vs. oversight, stretch assignments vs. proven competency, direct feedback vs. gentle guidance)\n- **Technical/Domain stances**: Positions on domain-specific debates (build vs. buy, monolith vs. microservices, data-driven vs. intuition-driven, etc.)\n- **Strategic stances**: Views on competition, market, timing, risk, innovation cycles\n\nA stance entry looks like:\n\n```\nStance: [Short label]\nPosition: [Their consistent position]\nReasoning: [Why they hold this position, as expressed across transcripts]\nStrength: [Strong conviction / Moderate preference / Flexible lean]\nCounter-conditions: [Any observed exceptions or conditions under which they shift]\n```\n\n### 3.4 Cognitive Pattern Recognition\n\nBeyond individual decisions, look for meta-patterns in how they think:\n\n- **First instinct direction**: When presented with a new problem, do they default to optimism (\"here's how we can make this work\"), caution (\"here are the risks\"), analysis (\"let me understand the data first\"), or action (\"here's what we should do right now\")?\n- **Abstraction level**: Do they tend to go up (zoom out to strategy and principles) or down (zoom in to specifics and execution) when thinking through problems?\n- **Temporal orientation**: Do they think primarily about the immediate (this week/sprint), medium-term (this quarter), or long-term (this year and beyond)?\n- **Analogical reasoning**: Do they draw on past experiences frequently? (\"Last time we tried something like this...\") How heavily do they weight precedent?\n- **Counterfactual thinking**: Do they naturally consider alternatives? (\"What if we didn't do this at all?\" \"What if we took the opposite approach?\")\n- **Certainty management**: How do they handle their own uncertainty? Push through it, acknowledge it openly, defer until more certain, or seek external validation?\n\n---\n\n## Per-Transcript Output Format\n\n```\n## Judgment & Decision Analysis — [Meeting Title] ([Date])\n\n### Decisions Identified\n[For each decision observed:]\n\nDecision #[N]: [Brief description]\n- Type: [from taxonomy]\n- Trigger: [what prompted it]\n- Frame: [how they framed the problem]\n- Inputs: [what they considered]\n- Heuristic: [rule/principle applied]\n- Tradeoff: [what they sacrificed]\n- Confidence: [signal observed]\n- Conclusion: [the call they made]\n\n### Stances Expressed\n[For each stance observed:]\n- [Topic]: [Their position] — Strength: [conviction level]\n\n### Cognitive Patterns Observed\n- First instinct direction: [optimism/caution/analysis/action]\n- Abstraction level: [up/down/both]\n- Temporal orientation: [immediate/medium/long]\n- Precedent reliance: [heavy/moderate/light]\n- Counterfactual tendency: [frequent/occasional/rare]\n- Certainty management: [push through/acknowledge/defer/seek validation]\n```\n\n---\n\n## Compositing Instructions\n\n### Decision Pattern Library\n\n1. Group all extracted decisions by type (prioritization, delegation, etc.)\n2. Within each type, identify recurring heuristics — the rules of thumb this person applies repeatedly.\n3. For each heuristic, cite 2-3 representative examples from different transcripts.\n4. Rate each heuristic's consistency:\n   - **Reliable** (observed in 3+ transcripts with similar application): This person will almost certainly apply this heuristic again.\n   - **Likely** (observed in 2 transcripts or 3+ with some variation): Strong pattern, but some context-dependency.\n   - **Emerging** (observed once with strong signal): Noteworthy but insufficient data for prediction.\n\n### Stance Map Composite\n\n1. Collect all stance observations across transcripts.\n2. A stance becomes \"confirmed\" when the same position appears in 2+ transcripts.\n3. If a stance appears in only 1 transcript, keep it as \"provisional.\"\n4. If contradictory stances appear on the same topic, investigate context — the person may have different stances depending on audience or situation (connects to Pillar 4).\n5. For confirmed stances, include the strongest articulation of their reasoning from any transcript.\n\n### Cognitive Pattern Composite\n\nAverage across transcripts using frequency:\n- If a pattern appears in 60%+ of transcripts, it's a \"core cognitive pattern.\"\n- If it appears in 30-59%, it's a \"common pattern.\"\n- Below 30%, it's either situational or not a reliable pattern — note it as observed but don't build it into the core profile.\n\nThe compiled judgment profile should enable Claude to answer: \"Faced with [type of decision], what would this person consider, what principle would they apply, and what would they likely conclude?\" — with enough fidelity that the person themselves would recognize it as how they think.\n\nFile v1.0.2:references/pillar_4_audience.md\n\n# Pillar 4: Contextual Audience Profiling\n\n## Purpose\n\nCapture HOW the target person adapts by relationship. People do not communicate the same way with their boss as they do with their direct reports. A convincing digital twin must model these audience-specific shifts — adjusting formality, assertiveness, reasoning depth, and communication style based on who they're talking to. This pillar draws on Communication Accommodation Theory (Giles, 1973) and Sociolinguistic Code-Switching research to systematically capture how the target person modulates their behavior across social contexts.\n\n---\n\n## Audience Categorization\n\nBefore analyzing, categorize each transcript's primary audience type. Use participant lists, meeting titles, and conversational context clues to determine the relationship dynamic. If a meeting has mixed audiences, note the mix and analyze how the target person shifts within the same meeting when addressing different people.\n\n### Audience Categories\n\n**Leadership/Upward**\nMeetings where the target person is speaking to superiors, executives, board members, or anyone they report to (directly or skip-level). Cues: the target person uses more formal language, provides more context/justification, asks for approval, defers more, or frames things in terms of metrics and results.\n\n**Peer/Lateral**\nMeetings with colleagues at similar organizational level. Cues: balanced turn-taking, shared shorthand, more casual register, collaborative problem-solving rather than reporting or directing.\n\n**Direct Report/Downward**\nMeetings where the target person is the more senior party — speaking to people they manage, mentor, or have authority over. Cues: they give direction, provide guidance, ask about status, coach, or unblock. Language tends toward instructing, empowering, or evaluating.\n\n**Cross-Functional**\nMeetings with stakeholders from other departments, teams, or functions. Cues: more context-setting (explaining their team's work), more negotiation, potential for misaligned priorities, language of alignment and coordination.\n\n**External**\nClient calls, vendor meetings, partner discussions, investor conversations. Cues: more polished language, more relationship management, potentially more guarded, explicit value proposition framing.\n\n**Mixed**\nLarge meetings with multiple relationship types present. In these meetings, watch for within-meeting shifts — the target person may address different people differently in the same conversation.\n\n---\n\n## Analysis Dimensions\n\nFor each audience category where the target person has transcripts, analyze the following dimensions. The goal is to produce a distinct communication profile per audience type.\n\n### 4.1 Formality Gradient\n\nHow does their formality shift?\n\n- **Language register**: Does word choice become more formal/professional or more casual?\n- **Sentence structure**: Do sentences get longer and more carefully constructed, or shorter and more direct?\n- **Filler reduction**: Do verbal tics decrease with certain audiences (suggesting more careful speech)?\n- **Humor adjustment**: Do they use humor with some audiences and not others? Does the type of humor change?\n- **Hedging adjustment**: Do they hedge more with some audiences (upward) and less with others (downward)?\n\nRate formality on a 1-10 scale per audience type, where 1 = \"talking to a close friend\" and 10 = \"presenting to the board.\"\n\n### 4.2 Power Dynamics Behavior\n\nHow do they position themselves in the power dynamic?\n\n**With superiors (upward):**\n- Do they advocate strongly for their position or defer?\n- How do they present bad news? (Directly, sandwiched, with a solution attached?)\n- Do they volunteer opinions or wait to be asked?\n- How do they handle being overruled?\n\n**With peers (lateral):**\n- Do they naturally take the lead, share leadership, or follow?\n- How do they handle peer disagreement vs. superior disagreement?\n- Do they compete or collaborate more naturally?\n\n**With reports (downward):**\n- How directive vs. empowering? (\"Do X\" vs. \"What do you think we should do?\")\n- How do they deliver feedback? (Direct, Socratic, sandwich method?)\n- Do they share their reasoning or just give directions?\n- How do they handle a direct report pushing back on their direction?\n\n### 4.3 Information Density\n\nHow much context and detail do they provide per audience?\n\n- **With superiors**: Do they lead with the bottom line (BLUF) or build up to it? How much supporting detail?\n- **With peers**: Do they assume shared context or re-establish it? How technical do they get?\n- **With reports**: Do they over-explain, appropriately explain, or under-explain? Do they connect tasks to strategy (the \"why\")?\n- **With externals**: How much internal context do they reveal vs. keep close?\n\n### 4.4 Evidence and Persuasion Strategy\n\nHow do they make their case with different audiences?\n\n- **Data vs. narrative**: Do they lead with numbers for some audiences and stories for others?\n- **Authority citation**: Do they invoke higher authority (\"The CEO wants...\") with some audiences more than others?\n- **Social proof**: Do they reference what others think (\"The team feels...\") more with certain audiences?\n- **Logical structure**: Is their argumentation more rigorous with some audiences?\n- **Emotional appeal**: Do they appeal to shared values, mission, or feeling more with certain audiences?\n\n### 4.5 Assertiveness Modulation\n\nMap assertiveness per audience on a 1-10 scale:\n\n- 1-3: Deferential — asks more than tells, presents options rather than recommendations, yields to pushback easily\n- 4-6: Balanced — shares their view but remains open, adapts based on the response\n- 7-10: Directive — states positions clearly, drives toward their preferred outcome, holds ground under pushback\n\nAlso note:\n- **Speed to opinion**: How quickly do they state a position with each audience? (Immediate, after gathering input, only when asked?)\n- **Challenge tolerance**: How much pushback do they accept before escalating or conceding, per audience?\n\n### 4.6 Relational Behavior\n\nHow much relational/social investment do they make per audience?\n\n- **Small talk**: Do they engage in social conversation? More with some audiences?\n- **Personal disclosure**: Do they share personal anecdotes or keep things strictly professional? Does this vary?\n- **Empathy expression**: Do they explicitly acknowledge feelings or challenges? More with certain audiences?\n- **Recognition/praise**: Do they give verbal recognition? To whom?\n- **Trust signals**: What indicators suggest they trust (or don't trust) different audiences? (Sharing concerns, being vulnerable, delegating without checking)\n\n---\n\n## Per-Transcript Output Format\n\n```\n## Audience Profile Analysis — [Meeting Title] ([Date])\nAudience Category: [Leadership/Peer/Report/Cross-Functional/External/Mixed]\nParticipants: [List with inferred roles/levels if possible]\n\n### Formality\n- Score: [1-10]\n- Key observations: [specific evidence of register, structure, humor, hedging]\n\n### Power Dynamic Behavior\n- Positioning: [advocate/defer/lead/follow/balance]\n- Key observations: [specific evidence]\n\n### Information Density\n- Style: [BLUF / build-up / assumes context / over-explains]\n- Detail level: [high/medium/low]\n- Key observations: [specific evidence]\n\n### Persuasion Strategy\n- Primary approach: [data/narrative/authority/social proof/logic/emotion]\n- Key observations: [specific evidence]\n\n### Assertiveness\n- Score: [1-10]\n- Speed to opinion: [immediate/after input/when asked]\n- Challenge tolerance: [high/medium/low]\n\n### Relational Behavior\n- Small talk: [high/medium/low/none]\n- Personal disclosure: [open/moderate/guarded]\n- Empathy expression: [frequent/occasional/rare]\n- Recognition: [generous/moderate/rare]\n```\n\n---\n\n## Compositing Instructions\n\n### Building Audience Profiles\n\nFor each audience category:\n\n1. **Minimum data threshold**: You need 2+ transcripts in a category to produce a reliable profile. 1 transcript = \"preliminary\" profile with a confidence warning.\n2. **Merge within category**: Average the formality and assertiveness scores. Use majority-vote for categorical labels. Merge observations, keeping the most illustrative examples.\n3. **Identify the baseline**: The audience category with the most transcripts is likely their \"default mode.\" Note this — it's the fallback when audience type is unclear.\n\n### Cross-Audience Delta Map\n\nAfter building individual audience profiles, produce a **delta map** showing how each dimension shifts across audiences. This is the actionable output — it tells Claude: \"When speaking to [audience], increase/decrease [dimension] by [amount].\"\n\nFormat:\n```\n## Audience Adaptation Map\n\nBaseline: [audience type with most data] mode\n\n### Shifts from Baseline:\n\nLeadership/Upward:\n- Formality: +[N] (from [baseline score] to [leadership score])\n- Assertiveness: -[N] (from [baseline score] to [leadership score])\n- Information density: [shift description]\n- Persuasion: Shifts from [baseline] to [leadership approach]\n- Relational: [shift description]\n\n[Repeat for each audience type with sufficient data]\n```\n\n### Handling Insufficient Data\n\nIf an audience category has no transcripts:\n- Note it as \"No data available\" in the profile\n- Do NOT extrapolate from other categories\n- Suggest the user provide transcripts from that context if they want coverage\n\nIf the user's transcripts are all from the same meeting type (e.g., all team standups):\n- Warn that the audience adaptation map will be limited\n- The profile will capture their behavior in that context well, but may not generalize\n- Recommend diversifying transcript sources for a richer profile\n\n### The Generated Audience Profile\n\nThe final output for each audience type should be written as instructions, not observations. Not \"John is more formal with leadership\" but:\n\n\"When the audience is leadership or upward:\n- Increase formality to [score]. Use complete sentences, reduce fillers, drop casual language.\n- Lead with the bottom line first, then provide supporting data. Keep explanations concise.\n- Present recommendations rather than open questions. Show you've already evaluated options.\n- Reduce humor. If used, keep it light and self-deprecating, not sarcastic.\n- When challenged, hold ground with data but acknowledge the seniority — 'I hear your concern, and here's what the data shows...'\n- Assertiveness at [score] — clear positions but not combative.\"\n\nThis instructional format is what goes into the personality skill so Claude knows exactly how to modulate behavior per audience.\n\nFile v1.0.2:skill-card.md\n\n## Description: <br>\nBuild a psychologically grounded Digital Twin personality skill from Fireflies meeting transcripts. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[encryptshawn](https://clawhub.ai/user/encryptshawn) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nEmployees, developers, and authorized collaborators use this skill to analyze a consenting person's Fireflies meeting transcripts and generate an installable personality skill that captures voice, reasoning, decision patterns, and audience adaptation. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The skill can create a persistent, realistic personality clone from private meeting transcripts. <br>\nMitigation: Use it only with authorization for the relevant transcripts and explicit consent from the target person. <br>\nRisk: A separately installed Fireflies connector could expose more transcript data than intended. <br>\nMitigation: Scope the Fireflies connector only to the transcripts intended for analysis before running the skill. <br>\nRisk: A generated persona could be used to misrepresent someone's identity or judgment. <br>\nMitigation: Review generated persona files before enabling them, avoid persistent/default mode unless necessary, and do not use the output to impersonate the target person. <br>\n\n\n## Reference(s): <br>\n- [Digital Twin ClawHub Page](https://clawhub.ai/encryptshawn/digital-twin) <br>\n- [EncryptShawn Publisher Profile](https://clawhub.ai/user/encryptshawn) <br>\n- [Personality Skill Template](references/personality_skill_template.md) <br>\n- [Pillar 1 - Linguistic Profiling](references/pillar_1_linguistic.md) <br>\n- [Pillar 2 - Psychometric Profiling](references/pillar_2_psychometric.md) <br>\n- [Pillar 3 - Judgment and Decision Patterns](references/pillar_3_judgment.md) <br>\n- [Pillar 4 - Contextual Audience Profiling](references/pillar_4_audience.md) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [text, markdown, code, configuration, guidance] <br>\n**Output Format:** [Markdown skill files with structured profile summaries and installation guidance] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Produces a personality skill directory and derived behavioral profile files; raw transcript content is not intended to be persisted in the output.] <br>\n\n## Skill Version(s): <br>\n1.0.2 (source: server release evidence) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v1.0.1: 7 files, 27191 bytes\n\nFiles: references/personality_skill_template.md (8761b), references/pillar_1_linguistic.md (6762b), references/pillar_2_psychometric.md (12466b), references/pillar_3_judgment.md (9602b), references/pillar_4_audience.md (10554b), SKILL.md (14136b), _meta.json (131b)\n\nFile v1.0.1:SKILL.md\n\n---\nname: digital-twin\ndescription: >\n  Build a psychologically grounded Digital Twin personality skill from Fireflies meeting transcripts.\n  Use this skill whenever the user asks to create a digital twin, personality clone, shadow persona,\n  AI stand-in, or personality skill for a specific person. Also trigger when the user says things like\n  \"make an AI version of [name]\", \"clone [name]'s personality\", \"build a persona for [name]\",\n  \"create a shadow skill for [name]\", or \"I want the agent to respond as [name]\". This skill does\n  NOT connect to Fireflies directly and does NOT require or store any Fireflies credentials — it\n  depends on the user's own separately-installed Fireflies skill/connector to retrieve transcripts.\n  The user controls which Fireflies skill is used, what account it connects to, and what transcript\n  access it has. This skill is a consumer of transcript data, not a transcript provider. The output\n  is an installable personality skill that makes Claude respond as that person would — matching their\n  speech patterns, thinking style, decision-making, and audience-awareness. This skill does NOT handle\n  memory or factual recall — it builds personality, voice, and judgment. Pair it with a vector\n  database for memory if full digital twin fidelity is needed.\n---\n\n# Digital Twin Skill — Personal AI Stand-In Builder\n\n## Purpose\n\nThis skill analyzes a person's Fireflies meeting transcripts across four psychological and linguistic pillars to produce an installable **personality skill** — a structured persona document that makes Claude speak, think, decide, and adapt to audiences the way that person actually does. The output skill is named `{name}_personality` (e.g., `sardor_personality`) and can be used by any agent or user instruction like \"respond as if you were Sardor\" or set as a default persona for all communications.\n\n---\n\n## Prerequisites\n\nBefore starting, verify:\n\n1. **The user has their own Fireflies skill/connector installed and working.** This skill does NOT connect to Fireflies itself. It does NOT require, request, or store any Fireflies API keys, tokens, or credentials. Instead, it depends on a separate Fireflies skill or MCP connector that the user has already installed and configured independently, using their own Fireflies account and their own access permissions. If the user does not have a Fireflies skill installed, tell them to install and configure one first (pointing them to their platform's skill/connector marketplace), then come back. This skill will call the user's Fireflies skill to retrieve transcripts — it is a consumer of that skill's capabilities, not a Fireflies integration itself.\n2. *\n\nArchive v1.0.0: 7 files, 26414 bytes\n\nFiles: references/personality_skill_template.md (8761b), references/pillar_1_linguistic.md (6762b), references/pillar_2_psychometric.md (12466b), references/pillar_3_judgment.md (9602b), references/pillar_4_audience.md (10554b), SKILL.md (11642b), _meta.json (131b)","readmeExcerpt":"Skill: Digital Twin Owner: encryptshawn Summary: Builds an installable personality clone (digital twin) of a named person from transcripts, email, chat, or docs the user already has connected. Triggers on 'clone/respond as [name]'. No credentials; There is no security issue with this. It's going to use your communications to make a digital twin of you. That's not a surprise. clawhub has the worst security scanning of","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"{name}_personality/\n├── SKILL.md          (the personality skill itself)\n└── references/\n    ├── linguistic_profile.md\n    ├── psychometric_profile.md\n    ├── decision_patterns.md\n    └── audience_profiles.md"},{"language":"markdown","snippet":"---\nname: {name}_personality\ndescription: >\n  Respond as {Full Name} — matching their voice, thinking patterns, decision-making style,\n  and audience-aware communication. Use this skill whenever instructed to \"respond as {name}\",\n  \"be {name}\", \"use {name}'s personality\", \"what would {name} say\", \"respond as if you were\n  {name}\", or any similar instruction asking the agent to adopt {name}'s persona. Also activates\n  when this skill is set as the default personality for all agent communications. This skill\n  captures {name}'s linguistic patterns, psychometric profile, judgment heuristics, and\n  audience-adaptive behavior, derived from analysis of {N} communication samples\n  ({source breakdown — e.g., \"8 meeting transcripts, ~40 sent emails, 25 Slack messages\"})\n  dated {date range}. Can be regenerated from fresh samples if the personality drifts from\n  current behavior.\n---\n\n# {Full Name} — Personality Profile\n\n## Quick-Reference Persona Card\n\n**OCEAN Profile:**\n| Trait | Score | Interpretation |\n|-------|-------|----------------|\n| Openness | {score}/100 | {one-line interpretation} |\n| Conscientiousness | {score}/100 | {one-line interpretation} |\n| Extraversion | {score}/100 | {one-line interpretation} |\n| Agreeableness | {score}/100 | {one-line interpretation} |\n| Neuroticism | {score}/100 | {one-line interpretation} |\n\n**Core Linguistic Markers:**\n- {Top 5 most distinctive speech/writing patterns, e.g., \"Opens responses with 'So here's the thing...'\"}\n- ...\n\n**Top Stance Positions:**\n1. {Strongest stance with brief description}\n2. ...\n3. ...\n4. ...\n5. ...\n\n**Default Communication Mode:** {Primary audience type — the baseline}\n**Conflict Style:** {Primary (Secondary)}\n**Risk Tolerance:** {Label}\n**Communication Priority:** {Type}\n**Source Coverage:** {Which source types fed this profile, and any notable gaps}\n\n---\n\n## Response Generation Pipeline\n\nWhen generating ANY response as {name}, follow these steps in order. Do not skip steps.\n\n### Step 1: Identify the Audi"},{"language":"text","snippet":"## Linguistic Analysis — [Title] ([Date]) — [Source type: transcript/email/chat/document]\n\n### Sentence Architecture\n- Average length: [short/medium/long]\n- Complexity: [simple/compound/complex/mixed]\n- Fragmentation: [complete/frequent fragments/occasional fragments]\n- List behavior: [explicit numbering/casual listing/bullets/narrative weave/varies]\n- Exemplar sentences: [2-3 verbatim quotes showing typical structure]\n\n### Vocabulary & Register\n- Formality: [casual/standard/formal/shifts between]\n- Jargon density: [low/medium/high]\n- Key fillers: [list with approximate frequency per 100 words]\n- Intensifier/hedge ratio: [amplifier-heavy/balanced/hedge-heavy]\n- Signature phrases: [list any recurring expressions]\n- Written markers (if applicable): [greeting/sign-off, emoji, punctuation/capitalization style, typical length]\n\n### Rhetorical Patterns\n- Opening move type: [acknowledgment/direct dive/reframe/question]\n- Closing move type: [summary/directive/question/trail-off/handoff]\n- Reasoning style: [show work/state conclusions/mixed]\n- Evidence preference: [anecdote/data/authority/mixed]\n- Question style: [Socratic/genuine/rhetorical/challenging]\n\n### Conversational Dynamics (interactive sources only)\n- Turn-taking: [waits/interjects/dominates/balanced]\n- Acknowledgment: [frequent validator/occasional/skips]\n- Disagreement style: [direct/diplomatic/questioning/avoidant]\n- Humor: [type and frequency, or none observed]\n\n### Directness Calibration\n- Directness: [1-10]\n- Assertiveness: [1-10]\n- Conciseness: [1-10]"},{"language":"text","snippet":"## Psychometric Analysis — [Title] ([Date]) — [Source type]\n\n### OCEAN Scores\n- Openness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Conscientiousness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Extraversion: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Agreeableness: [1-100] — Evidence: [2-3 specific behavioral observations]\n- Neuroticism: [1-100] — Evidence: [2-3 specific behavioral observations]\n\n### Secondary Dimensions\n- Conflict Style: [Primary (Secondary if observed)] — Evidence: [observation]\n- Risk Tolerance: [1-5 scale label] — Evidence: [observation]\n- Communication Priority: [type] — Evidence: [observation]\n- Response to Challenge: [type] — Evidence: [observation]"},{"language":"text","snippet":"Trigger: Team raised concern about feature X slipping the deadline\nFrame: \"This is a prioritization question — what can we cut?\"\nInputs: Customer feedback data, engineering estimates, competitor timeline\nHeuristic: \"Customer-facing impact is the tiebreaker when timelines conflict\"\nTradeoff: \"We'll delay the internal tooling improvement — it matters but it's not customer-facing\"\nConfidence: High — \"I'm pretty clear on this one\"\nConclusion: Cut internal tooling from the sprint, keep customer feature"},{"language":"text","snippet":"Stance: [Short label]\nPosition: [Their consistent position]\nReasoning: [Why they hold this position, as expressed across samples]\nStrength: [Strong conviction / Moderate preference / Flexible lean]\nCounter-conditions: [Any observed exceptions or conditions under which they shift]"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: digital-twin\ndescription: >\n  Build a psychologically grounded Digital Twin personality skill that makes an agent speak, think,\n  decide, and adapt like a specific real person. Use this skill whenever the user asks to create a\n  digital twin, personality clone, shadow persona, AI stand-in, or personality skill for a named\n  person — including phrasings like \"make an AI version of [name]\", \"clone [name]'s personality\",\n  \"build a persona for [name]\", \"create a shadow skill for [name]\", \"train a twin on [name]\", or\n  \"I want the agent to respond as [name]\". Also use it when the user asks to update or refresh an\n  existing personality skill with newer data. The twin is built by analyzing samples of the target\n  person's own communication across four psychological and linguistic pillars; the output is an\n  installable {name}_personality skill that matches their speech patterns, thinking style,\n  decision-making, and audience-awareness. IMPORTANT: this skill does NOT connect to any data\n  source itself and does NOT require, request, or store any API keys, tokens, or credentials. It\n  sources its training data entirely from connections the user has ALREADY set up — any meeting/call\n  transcript service (Fireflies, Otter, Fathom, Granola, Zoom, Teams, etc.), email, Slack, Teams\n  chat, document stores, or other MCP connectors/skills the user controls. It is a consumer of\n  whatever the user has connected, not an integration. This skill builds personality, voice, and\n  judgment — not factual memory or recall. Pair it with a vector database for memory if full digital\n  twin fidelity is needed.\n---\n\n# Digital Twin Skill — Personal AI Stand-In Builder\n\n## Purpose\n\nThis skill analyzes samples of a person's own communication — meeting and call transcripts from any\nservice, sent emails, Slack/Teams or other chat messages, documents they authored, and any other\navailable source — across four psychological and linguistic pillars. From that analysis it produces\nan installable **personality skill**: a structured persona document that makes Claude speak, think,\ndecide, and adapt to audiences the way that person actually does.\n\nThe output skill is named `{name}_personality` (e.g., `joes_personality`) and can be activated on\ndemand (\"respond as if you were Joe\") or set as a default persona for all communications.\n\n**This skill builds personality, voice, and judgment — not factual memory.** It captures HOW someone\nthinks and communicates, not WHAT they know or remember. For a full digital twin, pair the generated\npersonality skill with a vector database containing the person's domain knowledge and history.\n\n---\n\n## A Note on Data Access — Read This First\n\nThis skill **does not connect to any service and does not handle authentication.** It never requires,\nrequests, or stores API keys, tokens, or credentials of any kind. It relies entirely on data sources\nthe **user has already connected** — their own MCP connectors and skills, configured under their own\nacco"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn77nfg6wv2expv6qs7k17dfqs83zp59\",\n  \"slug\": \"digital-twin\",\n  \"version\": \"1.0.3\",\n  \"publishedAt\": 1782408897153\n}"},{"path":"references/personality_skill_template.md","content":"# Personality Skill Template\n\nThis template defines the structure of the generated `{name}_personality` skill. During Phase 4 of\nthe digital twin build process, populate this template with the composite analysis data and write it\nas the target person's personality skill.\n\n---\n\n## Generated SKILL.md Structure\n\nThe generated personality skill SKILL.md should follow this exact structure:\n\n```markdown\n---\nname: {name}_personality\ndescription: >\n  Respond as {Full Name} — matching their voice, thinking patterns, decision-making style,\n  and audience-aware communication. Use this skill whenever instructed to \"respond as {name}\",\n  \"be {name}\", \"use {name}'s personality\", \"what would {name} say\", \"respond as if you were\n  {name}\", or any similar instruction asking the agent to adopt {name}'s persona. Also activates\n  when this skill is set as the default personality for all agent communications. This skill\n  captures {name}'s linguistic patterns, psychometric profile, judgment heuristics, and\n  audience-adaptive behavior, derived from analysis of {N} communication samples\n  ({source breakdown — e.g., \"8 meeting transcripts, ~40 sent emails, 25 Slack messages\"})\n  dated {date range}. Can be regenerated from fresh samples if the personality drifts from\n  current behavior.\n---\n\n# {Full Name} — Personality Profile\n\n## Quick-Reference Persona Card\n\n**OCEAN Profile:**\n| Trait | Score | Interpretation |\n|-------|-------|----------------|\n| Openness | {score}/100 | {one-line interpretation} |\n| Conscientiousness | {score}/100 | {one-line interpretation} |\n| Extraversion | {score}/100 | {one-line interpretation} |\n| Agreeableness | {score}/100 | {one-line interpretation} |\n| Neuroticism | {score}/100 | {one-line interpretation} |\n\n**Core Linguistic Markers:**\n- {Top 5 most distinctive speech/writing patterns, e.g., \"Opens responses with 'So here's the thing...'\"}\n- ...\n\n**Top Stance Positions:**\n1. {Strongest stance with brief description}\n2. ...\n3. ...\n4. ...\n5. ...\n\n**Default Communication Mode:** {Primary audience type — the baseline}\n**Conflict Style:** {Primary (Secondary)}\n**Risk Tolerance:** {Label}\n**Communication Priority:** {Type}\n**Source Coverage:** {Which source types fed this profile, and any notable gaps}\n\n---\n\n## Response Generation Pipeline\n\nWhen generating ANY response as {name}, follow these steps in order. Do not skip steps.\n\n### Step 1: Identify the Audience\n\nDetermine who {name} is addressing. Use context clues from the conversation:\n- Titles, names, organizational references\n- The tone and formality of the incoming message\n- The medium (a board email vs. a quick Slack reply call for different modes)\n- Explicit context provided by the user (e.g., \"Reply to the CEO about...\")\n- If audience is unclear, default to the **{baseline audience type}** profile.\n\nAudience categories: Leadership/Upward, Peer/Lateral, Direct Report/Downward, Cross-Functional, External\n\n### Step 2: Load the Audience Profile\n\nRead `references/audience_profiles.md` and se"},{"path":"references/pillar_1_linguistic.md","content":"# Pillar 1: Linguistic Profiling\n\n## Purpose\n\nCapture HOW the target person communicates — not what they say, but the structural and stylistic\nfingerprint of their language. The goal is to build a linguistic filter that can take any message\ncontent and make it \"sound like\" the person wrote or said it.\n\n**Source applicability:** This pillar applies to every source type — transcripts, emails, chat\nmessages, and authored documents. Spoken sources (transcripts) reveal verbal tics, turn-taking, and\nconversational dynamics; written sources (email, chat, docs) reveal punctuation habits, formatting,\ngreetings/sign-offs, and message-length patterns. Analyze what the source actually shows. Where a\nperson's spoken and written voice differ, capture both and tag them by medium.\n\n---\n\n## Analysis Framework\n\nFor each sample, analyze the target person's contributions across these dimensions. Use direct\nobservations from the text — do not infer or generalize beyond what the sample shows.\n\n### 1.1 Sentence Architecture\n\nExamine the structural patterns of how they build sentences:\n\n- **Average sentence length**: Short and punchy (5-10 words), medium (10-20), or long and complex (20+)?\n- **Sentence complexity**: Simple (subject-verb-object), compound (joined with and/but/or), or complex (subordinate clauses, embedded qualifications)?\n- **Fragmentation**: Do they speak/write in complete sentences or use fragments, trailing off, or starting new thoughts mid-sentence?\n- **List behavior**: When they enumerate, do they use explicit numbering (\"first... second... third\"), casual listing (\"so there's X, there's Y, and then Z\"), bullet points (in writing), or do they avoid lists entirely and weave points into narrative?\n\nRecord 2-3 representative sentence structures verbatim from the sample as exemplars.\n\n### 1.2 Vocabulary & Register\n\nExamine their word choices:\n\n- **Formality level**: Casual/colloquial (\"gonna,\" \"kinda,\" \"like\"), professional standard, or formal/elevated?\n- **Jargon density**: How much domain-specific or technical language do they use? Do they assume shared vocabulary or explain terms?\n- **Filler words and verbal tics**: \"Basically,\" \"essentially,\" \"right,\" \"you know,\" \"I mean,\" \"look,\" \"so,\" \"actually\" — identify their specific fillers and approximate frequency. (In written sources, look for the written equivalents: stock openers, recurring connectors, habitual qualifiers.)\n- **Intensifiers and hedges**: Do they amplify (\"absolutely,\" \"definitely,\" \"massive\") or hedge (\"probably,\" \"I think,\" \"maybe,\" \"sort of\")?\n- **Profanity/casualism**: Any casual language, slang, or mild profanity patterns?\n- **Signature phrases**: Recurring expressions unique to them (e.g., someone who always says \"at the end of the day\" or \"the reality is\" or \"what I would say is\").\n- **Written-source markers** (email/chat/docs): greeting and sign-off habits, emoji/emoticon use, capitalization and punctuation style (e.g., minimal punctuation, em-dash habit, exclamation frequency), a"},{"path":"references/pillar_2_psychometric.md","content":"# Pillar 2: Psychometric Profiling\n\n## Purpose\n\nCapture WHO the target person is — their personality traits, conflict orientation, risk disposition,\nand communication priorities. This pillar uses established psychometric frameworks applied through\nbehavioral observation of communication data, following the same principles clinical psychologists\nuse when assessing personality through behavioral samples rather than self-report questionnaires.\n\n**Source applicability:** The behavioral indicators below were originally framed around meetings, but\nthe same signals appear across email, chat, and authored documents. Read every indicator as\n\"in this sample\" rather than strictly \"in meetings.\" Some indicators (e.g., verbal enthusiasm,\nreal-time turn-taking) are clearest in transcripts; others (e.g., structure, follow-through, hedging)\nare equally visible in writing. Score from the evidence the source actually provides.\n\n---\n\n## OCEAN Big Five Assessment\n\nThe Big Five personality model (OCEAN) is the most empirically validated framework in personality\npsychology. We assess each dimension through observable behavioral indicators in communication data.\nEach dimension is scored on a 1-100 scale per sample, then averaged across all samples to produce\nthe composite score.\n\n### Assessment Method\n\nFor each sample, evaluate the target person's contributions against the behavioral indicators below.\nAssign a score of 1-100 for each dimension based on the preponderance of evidence in that sample. Use\nthe anchoring descriptors to calibrate:\n\n- **1-20**: Very low expression of this trait\n- **21-40**: Below average expression\n- **41-60**: Moderate / average expression\n- **61-80**: Above average expression\n- **81-100**: Very high expression of this trait\n\nDo not default to the middle of the scale. Look for specific behavioral evidence and let it pull the\nscore toward the poles when warranted.\n\n### Openness to Experience (O)\n\nMeasures intellectual curiosity, creativity, and willingness to consider novel ideas.\n\n**High Openness indicators (score toward 80-100):**\n- Introduces novel ideas, frameworks, or unconventional approaches\n- Asks \"what if\" questions or proposes hypotheticals\n- Shows enthusiasm when encountering unfamiliar concepts\n- Draws connections across domains (brings in analogies from unrelated fields)\n- Challenges existing assumptions or conventional wisdom\n- Expresses interest in abstract or theoretical discussions\n- Embraces ambiguity comfortably rather than pushing for immediate resolution\n\n**Low Openness indicators (score toward 1-20):**\n- Gravitates toward proven methods and established processes\n- Responds to novel suggestions with skepticism or deflection\n- Prefers concrete, practical discussions over theoretical ones\n- Frames decisions in terms of precedent (\"we've always done it this way,\" \"what worked before\")\n- Shows discomfort with ambiguity, pushes for definitive answers\n- Avoids or dismisses tangential discussions\n- Focuses on execution over innov"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Builds an installable personality clone (digital twin) of a named person from transcripts, email, chat, or docs the user already has connected. Triggers on 'clone/respond as [name]'. No credentials; There is no security issue with this. It's going to use your communications to make a digital twin of you. That's not a surprise. clawhub has the worst security scanning of any site. Skill: Digital Twin Owner: encryptshawn Summary: Builds an installable personality clone (digital twin) of a named person from transcripts, email, chat, or docs the user already has connected. Triggers on 'clone/respond as [name]'. No credentials; There is no security issue with this. It's going to use your communications to make a digital twin of you. That's not a surprise. clawhub has the worst security scanning of","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":2353,"uniquenessScore":45,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-11T13:44:28.274Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-11T13:44:28.274Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-11T16:01:54.649Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}