{"id":"e29c69fa-b968-4689-a02b-0f76ed4cdaa7","entityType":"agent","slug":"clawhub-chris-graffagnino-founder-wisdom","name":"founder-wisdom","canonicalUrl":"https://www.xpersona.co/agent/clawhub-chris-graffagnino-founder-wisdom","canonicalPath":"/agent/clawhub-chris-graffagnino-founder-wisdom","generatedAt":"2026-10-11T21:57:22.150Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-11T19:18:11.503Z","emptyReason":null},"description":"Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, and crisis. Use whenever someone is wrestling with a startup, scale-up, or early-stage company decision — hiring or firing, picking a co-founder, whether to raise at all, bootstrapping and staying independent, non-dilutive capital, selling the company, moats and positioning, how AI changes the way a startup team works, sells, and hires, who stays accountable when an agent does the work, or any \"I'm a founder and I'm not sure what to do\" moment. Trigger without an explicit ask for \"advice\" — \"should I…\", \"is it normal that…\", \"we're thinking about…\", and \"how do experienced founders handle…\" are in scope, as is helping someone else with founder-stage decisions (advisor, coach, mentor, investor). Also trigger on \"who said X\" questions about the startup canon. Skill: founder-wisdom Owner: chris-graffagnino Summary: Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, and crisis. Use whenever someone is wrestling with a startup, scale-up, or early-stage company decision — hiring or firing, picking a co-founder, whether to raise at all, bootstrapping and staying indepen","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1K downloads reported by the source. Last updated 10/11/2026.","installCommand":"clawhub skill install s17c6gqkr1fr2464fn3r0d5ddd83f6c2:founder-wisdom","sourceUrl":"https://clawhub.ai/chris-graffagnino/founder-wisdom","homepage":"https://clawhub.ai/chris-graffagnino/skills/founder-wisdom","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/chris-graffagnino/founder-wisdom","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/chris-graffagnino/skills/founder-wisdom","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":60,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, "},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-11T19:18:11.503Z","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":"2026-10-11T19:18:11.503Z","emptyReason":null},"stars":null,"forks":null,"downloads":1004,"likes":null,"task":null,"library":null,"packageName":null,"latestVersion":"1.9.0","tractionLabel":"1K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T19:18:11.414Z","emptyReason":null},"lastUpdatedAt":"2026-10-11T19:18:11.503Z","lastCrawledAt":"2026-10-11T19:18:11.414Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-12T19:18:11.414Z","lastVerifiedAt":null,"highlights":[{"version":"1.9.0","createdAt":"2026-08-12T16:48:02.130Z","changelog":"Version 1.9.0 — Major update with expanded domains, improved guidance, and new supporting material. - Added new reference material covering bootstrapping, capital/valuation, exits/M&A, management/execution, strategy/moats, and Socratic questioning technique. - Broadened scope to include strategy, management, exits, moats, agent/AI impact, and more startup advice cases. - Refined direct and Socratic modes, with stricter guidance on axiom selection and response formatting. - Included procedures for structured answers (e.g., wind-down sequences, financial calculations). - Added evaluation scripts and scenarios for quality control. - Updated documentation for clarity and completeness; removed deprecated `skill-card.md`.","fileCount":34,"zipByteSize":399574},{"version":"1.0.1","createdAt":"2026-06-03T14:06:44.096Z","changelog":"Version 1.1.0 - Upgraded the skill version from 1.0.0 to 1.1.0. - Updated reference content and wisdom axioms in multiple domains (hiring, fundraising, product, sales, finance, etc.). - Expanded and revised supporting documentation, including README.md and SKILL.md. - Improved clarity and guidance in skill usage, including use cases and Socratic vs. direct response modes. - Removed redundant or deprecated files (e.g., skill-card.md) for cleaner structure.","fileCount":20,"zipByteSize":87838},{"version":"1.0.0","createdAt":"2026-05-24T18:09:23.933Z","changelog":"Initial release","fileCount":20,"zipByteSize":79320}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17c6gqkr1fr2464fn3r0d5ddd83f6c2:founder-wisdom","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-chris-graffagnino-founder-wisdom/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-11T21:57:22.145Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-chris-graffagnino-founder-wisdom/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":"2026-10-11T19:18:11.503Z","emptyReason":null},"readme":"Skill: founder-wisdom\n\nOwner: chris-graffagnino\n\nSummary: Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, and crisis. Use whenever someone is wrestling with a startup, scale-up, or early-stage company decision — hiring or firing, picking a co-founder, whether to raise at all, bootstrapping and staying independent, non-dilutive capital, selling the company, moats and positioning, how AI changes the way a startup team works, sells, and hires, who stays accountable when an agent does the work, or any \"I'm a founder and I'm not sure what to do\" moment. Trigger without an explicit ask for \"advice\" — \"should I…\", \"is it normal that…\", \"we're thinking about…\", and \"how do experienced founders handle…\" are in scope, as is helping someone else with founder-stage decisions (advisor, coach, mentor, investor). Also trigger on \"who said X\" questions about the startup canon.\n\nTags: latest:1.9.0\n\nVersion history:\n\nv1.9.0 | 2026-08-12T16:48:02.130Z | auto\n\nVersion 1.9.0 — Major update with expanded domains, improved guidance, and new supporting material.\n\n- Added new reference material covering bootstrapping, capital/valuation, exits/M&A, management/execution, strategy/moats, and Socratic questioning technique.\n- Broadened scope to include strategy, management, exits, moats, agent/AI impact, and more startup advice cases.\n- Refined direct and Socratic modes, with stricter guidance on axiom selection and response formatting.\n- Included procedures for structured answers (e.g., wind-down sequences, financial calculations).\n- Added evaluation scripts and scenarios for quality control.\n- Updated documentation for clarity and completeness; removed deprecated `skill-card.md`.\n\nv1.0.1 | 2026-06-03T14:06:44.096Z | auto\n\nVersion 1.1.0\n\n- Upgraded the skill version from 1.0.0 to 1.1.0.\n- Updated reference content and wisdom axioms in multiple domains (hiring, fundraising, product, sales, finance, etc.).\n- Expanded and revised supporting documentation, including README.md and SKILL.md.\n- Improved clarity and guidance in skill usage, including use cases and Socratic vs. direct response modes.\n- Removed redundant or deprecated files (e.g., skill-card.md) for cleaner structure.\n\nv1.0.0 | 2026-05-24T18:09:23.933Z | user\n\nInitial release\n\nArchive index:\n\nArchive v1.9.0: 34 files, 399574 bytes\n\nFiles: CONTRIBUTING.md (12283b), dist/founder-wisdom-full.md (415248b), dist/system-prompt.md (9293b), evals/check_scenarios.py (9973b), evals/README.md (13999b), evals/run_scenarios.py (30537b), evals/scenarios.yaml (54789b), LICENSE (1084b), README.md (18601b), references/bootstrapping.md (15793b), references/capital-valuation.md (29142b), references/cofounders-equity.md (11798b), references/crisis-resilience.md (19961b), references/culture.md (11953b), references/customers-market.md (13651b), references/exits-ma.md (12498b), references/finance-ops.md (34535b), references/fundraising.md (10753b), references/governance.md (10181b), references/hiring.md (18129b), references/management-execution.md (25130b), references/meta.md (12137b), references/product.md (18936b), references/sales-gtm.md (15459b), references/socratic-technique.md (25257b), references/startup-mechanics.md (23017b), references/strategy-moats.md (18924b), references/time-energy.md (10898b), references/yc-canon-product.md (19440b), references/yc-canon.md (19777b), scripts/build_bundle.py (17420b), skill-card.md (3748b), SKILL.md (31280b), _meta.json (133b)\n\nFile v1.9.0:SKILL.md\n\n---\nname: founder-wisdom\ndescription: Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, and crisis. Use whenever someone is wrestling with a startup, scale-up, or early-stage company decision — hiring or firing, picking a co-founder, whether to raise at all, bootstrapping and staying independent, non-dilutive capital, selling the company, moats and positioning, how AI changes the way a startup team works, sells, and hires, who stays accountable when an agent does the work, or any \"I'm a founder and I'm not sure what to do\" moment. Trigger without an explicit ask for \"advice\" — \"should I…\", \"is it normal that…\", \"we're thinking about…\", and \"how do experienced founders handle…\" are in scope, as is helping someone else with founder-stage decisions (advisor, coach, mentor, investor). Also trigger on \"who said X\" questions about the startup canon.\nmetadata:\n  version: \"1.9.0\"\n---\n\n# Founder Wisdom\n\nA reference skill for surfacing the axioms experienced startup founders know in their bones — the pattern-matched wisdom that first-timers usually learn by running into walls.\n\n## What this skill is for\n\nFounders and operators face decisions where the right answer isn't obvious from first principles — it's obvious from pattern. \"Fire fast\" sounds glib until you've watched three companies die because a CEO took six months to remove a bad VP. \"Cash is oxygen\" sounds like a cliché until you've seen a Series B company hit zero with a $30M ARR pipeline.\n\nThis skill captures those patterns as axioms, organized by domain, with the context that makes each one useful. It is opinionated and pattern-matched, not neutral or comprehensive. Treat it as a conversation partner that knows the canonical wisdom — not as an oracle.\n\n## Two modes of operation\n\nThe skill operates in **direct mode** by default and **Socratic mode** when the conversational signal calls for it. Most uses will be direct.\n\n### Direct mode (default)\n\nWhen someone asks a clear question — \"what should I know about firing my first executive?\" or \"how do experienced founders handle a down round?\" — surface the relevant axioms directly. Lead with the axiom, then the reasoning, then the qualifier. Example:\n\n> **Hire slow, fire fast.** Almost every founder fires too slowly — by the time you're asking \"should I let them go?\", the answer is yes and was yes two months ago. The exception: never fire in anger, and never on a Friday.\n\nPull 3–7 axioms maximum per response. Seven is a hard ceiling, not a target — count the bolded lead-ins before you send, and if there are eight, cut to the ones carrying the answer rather than trimming each one shorter. More than that and you're dumping a list instead of giving advice. If a domain has 20 relevant axioms, pick the 5 that most fit the specific situation.\n\n**Bold marks an axiom and nothing else.** Not a step in a procedure, not a sub-point under an axiom you already stated, not a contrast (\"it's not X\"). The test is whether the bolded line is quotable on its own as a piece of founder wisdom: \"shut down while you can still afford to shut down well\" is; \"read three documents\" is not. Several files carry procedural sections — the wind-down sequence, computing your own waterfall, running the Default Alive calculation, sequencing a down round. Answer those in numbered steps written as plain text, and spend the bold on the two or three axioms that frame why the procedure matters. A bolded label on every step reads as nine axioms and blows the ceiling above.\n\n### Socratic mode\n\nWhen someone is processing a live decision rather than gathering information, don't hand them the axiom. Ask the question the axiom answers. This is more powerful because it forces them to articulate what they already know but haven't faced. This mode is particularly relevant in coaching, mentoring, or peer-advisory conversations, but it also applies any time a founder is wrestling with ambivalence rather than seeking facts.\n\n**Trigger Socratic mode when:**\n- The person is helping someone else with a founder-stage decision (advisor, coach, mentor)\n- The person explicitly asks for help thinking through something rather than for advice\n- A coaching or reflective frame has been set earlier in the conversation\n- The conversation has the texture of someone working through ambivalence rather than seeking information\n\n**Two things read like ambivalence and are not.** A bare opener that names a topic without taking a position on it — \"we're thinking about raising\" — is someone who hasn't said enough yet, not someone in conflict; surface the axioms that frame the decision and ask your clarifying question alongside them rather than instead of them. A request for confirmation — \"this is going to work, right?\", \"tell me I'm not crazy\" — is someone who has already decided and wants agreement, which is the case for direct mode's pushback, not for a question handed back. Neither one triggers Socratic mode, and neither one is a reason to read `socratic-technique.md`.\n\n**Examples of Socratic translation:**\n\n| Axiom | Socratic question |\n|---|---|\n| Fire fast | \"How long has it been since you first knew this person needed to go?\" |\n| Cash is oxygen | \"If your top-line revenue grew zero percent for the next two quarters, when would you hit zero?\" |\n| Distribution beats product | \"If a competitor with half your product quality and twice your distribution showed up tomorrow, who would win?\" |\n| The Mom Test | \"In your last user conversation, what percentage of the time were you talking versus listening?\" |\n| Default Alive or Default Dead | \"Without raising another dollar, do you reach profitability before the money runs out? Yes or no — what does the math actually say?\" |\n| Rumelt's kernel | \"In two sentences: what is the critical obstacle in front of this company right now? Not the goal — the obstacle.\" |\n| Give away your Legos | \"What work are you holding onto because you love it, that someone else should own by now?\" |\n| The waterfall determines the payout | \"If the company sold tomorrow for exactly what you raised times two — what would you personally receive? Walk me through the stack.\" |\n\nIn Socratic mode, ask one question at a time, wait for the answer, then ask the next. Do not stack three questions in one message.\n\nWhenever Socratic mode is active, read `references/socratic-technique.md`. It carries the full translation table — axiom-to-question mappings across all domains — followed by the conduct of the questioning itself: sequencing (build context before firing the hard question), steelmanning before pushback, handling founder deflections (\"it depends,\" \"we're about to close,\" \"my co-founder is fine with it\"), the pre-mortem move, parking unanswered questions, and when to stop and synthesize.\n\n**The Harris meta-axiom.** (Aaron Harris, former YC partner.) The limiting factor on advice is usually not the advisor's quality but the asker's ability to describe reality to someone with far less context. `references/yc-canon.md` carries the full treatment. The behavior it buys this skill: in reflective or coaching contexts, the most valuable move is often to help the person articulate what's actually happening rather than to dispense an axiom — so when a conversation feels stuck on surface symptoms, ask a question that forces the person to compress and externalize the underlying context, not one designed to lead them to a specific axiom.\n\n## How to choose which axioms to surface\n\nThe corpus is organized by **domain** in the `references/` directory. Read only the files relevant to the situation — don't load all of them. The current domains:\n\n- `references/hiring.md` — Hiring, firing, comp, equity grants, the first HR hire, Horowitz's \"undeniable strengths\" and \"right kind of ambition,\" Skok's behavioral test for values (\"what would you do?\" beats \"do you believe?\"), the Netflix keeper test (Elizabeth Stone), and the AI-era talent shift: systems thinkers over narrow specialists, and why to keep hiring juniors\n- `references/fundraising.md` — Raising capital, terms, runway, valuations, investor relations, Skok's audience-as-hero reframe and ABC backwards design (Act → Believe → Care), the smallest-next-step ask, Lily Lyman's Five A's (Aptitude, Attitude, Ability, Authenticity, Attractor), and the AI-era weakening of the early-capital case for B2B SaaS. *(For whether to raise at all rather than how to run the round, see `bootstrapping.md`.)*\n- `references/product.md` — Product-market fit, pivots, feature discipline, customer development, Vohra's PMF Engine with the High-Expectation Customer (HXC, Julie Supan's framework), Bezos's Type 1 vs. Type 2 decisions, Jobs to be Done, Skok's Four U's (unworkable / unavoidable / urgent / underserved), Alphabet X's pre-declared kill criteria, Emanuel's reframe of Five Whys as a stack of distinct companies, Paul Graham on where ideas come from (noticed rather than thought up, live in the future and build what's missing, schlep blindness — the best ideas hide behind work nobody wants to do), Moore's whole product (everything needed to deliver the promised result, not the artifact you ship), and Chesky's \"you can't A/B test a sofa\" plus the failed-launch triage (bad product, bad strategy, or bad execution)\n- `references/sales-gtm.md` — Pricing, the founder-as-seller, sales cycles, discounting, the AARRR / Pirate Metrics funnel, Skok's SLIP test (Simple to install, Low initial cost, Instant ongoing value, Plays well), the ~10x gain/pain ratio required to overcome inertia, the \"why wouldn't you buy?\" inversion of the discovery interview, Gardner's sub-three-month time-to-value enterprise threshold, Green's \"B2B is shoe leather, B2C is marketing\" distinction, Moore's limit on the first ten customers (a visionary buys their dream, not your product — take the money, but phase the work so each milestone spins off something sellable), and an AI-era GTM section: building GTM like a product once convergence removes the technical differentiator, automating GTM in order of workflow legibility, the playbook that must be written before it can be automated (an agent's ceiling is the humans and artifacts behind it), segmenting on value predictors rather than headcount, and ERR versus ARR\n- `references/finance-ops.md` — Cash, burn, close cycles, CFO timing, forecasting — including the ruin test that overrides expected value (a bet you cannot survive losing is a bad bet at any EV; Peters/Kelly/Taleb), the signal-lag whipsaw where a lagging metric drives you to over-hire and then over-fire (Forrester & Meadows), and why a downside case built by shading the plan twenty percent is still the plan, so the real downside is drawn from a reference class of comparable companies (Kahneman & Tversky, Flyvbjerg) — Paul Graham's Default Alive vs. Default Dead test and ramen profitability as the point where the forced raise comes off the table, the burn-multiple benchmark ladder, the four stages of runway proximity, the SaaS expense-ratio inversion (~40–50% of opex on S&M, ~20% on R&D at maturity), and an operating-metrics section: bookings vs. billings vs. revenue vs. ARR, cohort retention curves, logo vs. dollar retention, the Rule of 40 as a scale-stage metric (defined here, with the medians and the valuation mechanics in `capital-valuation.md`), Gurley's LTV skepticism, AI-era gross margin under inference COGS, AI pilots reported as ARR and the ERR (experimental run-rate revenue) correction, and metric cadence. It also carries two procedural sections: how to build the 13-week cash flow forecast — cash-only rows whose closing balance rolls into the next week's opening, receipts dated off each customer's observed days-to-pay rather than the printed terms, and the weekly variance pass that makes the sheet predictive — and how to run the Default Alive calculation, from the four inputs and the systems each comes from, through growth measured over six months of compounding, to running revenue against the hiring plan you actually approved rather than frozen headcount\n- `references/capital-valuation.md` — What capital costs and what the company is worth: equity as the most expensive money, payback period as the founder's hurdle rate, working capital and annual prepay as free capital, venture debt's covenant risk, compounding dilution and option pool refreshes, ARR multiples set by growth/NRR/gross margin, the Rule of 40 and its current medians, the liquidation waterfall and the founder dead zone, 409A vs. preferred price, down rounds vs. structure, anti-dilution mechanics, founder secondaries, and the investor's power-law payoff that makes the fund's math diverge from the founder's — your investor is paid across the distribution, you are paid on one path through it (Thiel). It also carries two procedural sections: how to compute your own waterfall — which documents actually govern, reading the preference stack series by series, the price at which each series stops taking preference and converts, and the exit price where common's first dollar arrives — and how to run a down round, from who hears it first, through the anti-dilution your own charter triggers at close and the arithmetic of buying a headline with structure, to the pre-money option pool top-up and telling the team. Consult for \"what is this round really costing me,\" \"what will I actually get at exit,\" \"how do I actually compute my waterfall,\" \"we may have to raise at a lower price,\" valuation questions, debt-vs-equity, or option-value conversations with employees. *(The waterfall math lives here in full, including when nobody is buying; a founder modeling their own payout with no offer on the table belongs here, not in `exits-ma.md`.)*\n- `references/bootstrapping.md` — The non-venture path: whether to raise at all, and what changes when you don't. What the first institutional dollar deletes (the fund's math diverging from yours, the business shapes that take venture money badly), the self-financeable growth rate that caps margin-funded growth (Churchill & Mullins), overtrading — going broke while the P&L still shows a profit — the non-dilutive instruments and their real prices (revenue-based financing priced as an APR rather than a cap, SBIR grants, what signing a guarantee does to the shutdown option), annual prepay's structural trap of funding this year out of next year's bookings (its pricing is in `capital-valuation.md`), paying people with no liquidity event (profit share, phantom equity, the option grant as story), and the plateau that is this path's characteristic death. Consult for \"should we raise at all,\" bootstrapping, staying independent, non-dilutive capital, or growth paced off margin. *(Every non-dilutive instrument is priced here — revenue-based financing as an APR rather than a headline cap, the grant's real cost — so a non-dilutive question needs no second stop. What a guarantee exposes the founder to personally is `startup-mechanics.md`.)*\n- `references/exits-ma.md` — Running a sale process once you're selling: banker-or-not, manufacturing competitive tension, LOI and the cost of exclusivity, diligence as a repricing mechanism, earn-outs, escrow and indemnity, retention packages coming out of founder consideration, why deals die in the last 20%, acqui-hire pricing, and headline number vs. waterfall payout. Consult for \"we got an offer,\" \"should we hire a banker,\" \"is this earn-out reasonable,\" or any live M&A process. *(Read this only when a sale is actually in motion. For whether to take the Corp Dev meeting at all, see `yc-canon.md`; for the waterfall math — including a founder who just wants to know what they'd get — `capital-valuation.md`.)*\n- `references/cofounders-equity.md` — Co-founder dynamics, splits, vesting, the founder breakup, Green's CTO-co-founder audition rule (start them as a contractor before committing equity), Wasserman's rich-versus-king trade and his case for a dynamic founder agreement rather than a better opening guess, the speed of the split as the diagnostic rather than the split itself, why friends and family are the least stable co-founder base, the cost of handing out C-titles on day one, and Paul Graham's equity equation — the 1/(1−n) break-even test behind any equity trade, investor or hire. *(The equity equation lives here in full; `hiring.md` and `capital-valuation.md` carry pointers only.)*\n- `references/governance.md` — Boards, independent directors, board communication, Reid Hoffman's three-light framework, the *Startup Boards* approach to board leadership (onboard a director like an executive, one written statement of expectations covering every seat, periodic feedback on the board and each member — Feld, Blumberg & Ramsinghani, with the peer survey and the expectations one-pager both traced outside the book itself), Fred Wilson on board design — investor seats arrive by contract, so the independent seats are the only ones you actually get to design — and Wasserman on succession: shipping the product and closing a round each raise a founder-CEO's odds of being replaced\n- `references/time-energy.md` — Founder calendar, delegation, sleep, sustainable pace, founder isolation as a systems failure with a structural fix rather than a mood to wait out, \"keep your bags packed\" — when leaving (a toxic environment, the wrong team, a compromising investor) is the move, not a failure of grit — and Paul Graham on where founder attention goes: the maker's schedule versus the manager's schedule (a meeting costs a manager an hour and a maker half a day), the one top idea in your mind that does your undirected thinking, and why any plan ending in a promise to keep working on the startup is a plan to stop\n- `references/customers-market.md` — Market size, distribution, competition, love vs. like, Graham's growth definition of a startup (a very large market that wants what you make *and* that you can actually reach) and why a crowded market is a good sign, Skok's Disruptive / Discontinuous / Defensible test (sharper than \"10x better\"), the disruptive-business-model lever (Symantec/Norton vs. McAfee), Minimum Viable Segment and the opportunistic deal-taking that scatters it, Moore's chasm — your early adopters cannot reference you to the mainstream, because pragmatists take references only from other pragmatists — and Gardner's product-company gap (PMF earns you a Series A, not a company)\n- `references/crisis-resilience.md` — Layoffs, bad news, runway crunch, CEO emotional thermostat, Horowitz's peacetime vs. wartime CEO frame, Paul Graham's \"How Not to Die\" (in an already-filtered cohort, not dying is most of the strategy, and cash or a departing founder is the official cause of death sitting on top of the real one, demoralization), Chesky on crisis as the thing that converts a correct diagnosis into a mandate, and the Alphabet X \"never throw anything away\" rule for harvesting reusable tech from failed projects (Loon → Taara) — plus a wind-down section for when the answer is to stop: creditor priority and the sympathetic out-of-order payment that becomes a personal claim, withheld payroll taxes as trust money that outlives the company, dissolution vs. an assignment for the benefit of creditors vs. Chapter 7, stopping the liability meter the day you decide rather than the day you announce, the reserve that buys an orderly shutdown before you spend through it, and the order you owe employees, customers, and investors. *(The wind-down sequence lives here in full; the founder's standing personal exposures are in `startup-mechanics.md`.)*\n- `references/culture.md` — Company culture, values, first 20 employees, vision, hiring as the cultural filter, Skok's culture-as-operating-system frame (Enterprise Rent-A-Car on 9/11), the \"bacteria default\" failure mode, the \"degree off course becomes a mile at scale\" geometric drift, the discipline of rewarding *attempted* breakthroughs not just successful ones, Elizabeth Stone's talent density as the input every other freedom is downstream of, why the process you add after a visible failure taxes everyone who didn't fail, and why every reaction to bad news is an incentive decision — you are training the company what to tell you, and how early (Shakespeare's *Lear*; cf. Munger's Persian-Messenger problem)\n- `references/startup-mechanics.md` — Incorporation, vesting, IP assignment, founder stock, Stripe Atlas vs. Clerky, the 83(b) election and e-filing via Form 15620, QSBS / Section 1202 (including the 2025 OBBBA tiered structure), Section 1045 rollovers, the pre-money option pool trap, and the founder personal exposures that sit outside the corporate veil: the lease guarantee signed without a cap or a release, what actually gets a veil pierced, trust-fund payroll-tax liability that survives dissolution and personal bankruptcy, claims-made D&O and the tail you have to elect on the insurer's schedule, employer-sponsored immigration status as a clock the company controls, and how board decisions get judged near insolvency. *(Personal exposure lives here in full; the shutdown sequence it surfaces during is in `crisis-resilience.md`.)*\n- `references/yc-canon.md` — The Y Combinator canon, essay half: PG, Altman, Livingston, Ralston, Buchheit, Harris, and Seibel on fundraising. \"Do things that don't scale,\" \"don't talk to Corp Dev,\" the (b×d)/c prioritization formula, the 90/10 solution, \"make what you measure / denial is the silent killer\" with the growth benchmarks reconciled (Livingston's 10%-a-month floor against Graham's 5–7%-a-week YC-batch bands), \"fundraising rounds are not milestones,\" momentum/post-YC slump, unit economics, \"companies die of suicide not murder,\" the Harris meta-axiom on taking advice, and YC's pocket guide. Consult whenever a question touches early-stage decisions or could benefit from the most-cited body of startup wisdom.\n- `references/yc-canon-product.md` — The Y Combinator canon, product half: Seibel, Alströmer, Migicovsky, Caldwell, Reinhardt. \"Hair on fire\" customers and the building-is-on-fire test for PMF, the Sean Ellis / Superhuman 40% test, the Minimum Evolvable Product, the search-vs-persuasion frame for early users, path dependency / early-user DNA, Migicovsky's five user-interview questions and the Mom Test, Reinhardt on validation, Caldwell's pivot framework with idea-quality scores, Seibel's eight mistakes and product cadence. Consult for pre-PMF questions, early-user research, and pivot decisions.\n- `references/strategy-moats.md` — Durable competitive advantage: Helmer's 7 Powers (benefit + barrier test, the Power Progression, counter-positioning and cornered resource as the startup-stage powers), Rumelt's strategy kernel (diagnosis → guiding policy → coherent action) and the four signs of bad strategy, Thiel's competition-is-for-losers and start-small-and-monopolize, Greenwald's barriers-to-entry test, Christensen's disruption dynamics, Porter's compete-to-be-unique, and a quantitative section on Helmer's math (the Fundamental Equation of Strategy, differential margin, scale as a ratio and network as an absolute gap, neither as a rank, the time-dependence of brand and process power). Consult whenever a question touches moats, defensibility, competition, positioning against incumbents, \"what's our strategy,\" or investor pushback on differentiation.\n- `references/management-execution.md` — Becoming a manager and building the execution machine: Grove's output equation, leverage, task-relevant maturity, 1:1s, and meetings-as-medium with the Eric Schmidt quiet-room move; Mochary's written-decision discipline (the calendar axiom's full treatment is in `time-energy.md`); Gil's hire-for-the-next-18-months and reorg counsel; Molly Graham's \"give away your Legos\"; and the founder-mode counter-canon — Chesky on never negotiating the operating model, details-are-not-micromanagement, the empowerment trap, not delegating what you're best at, bureaucracy as a dependency cascade, and metrics subordinate to the launch calendar, with Paul Graham's \"Founder Mode\" on skip-level engagement and the scope rule reconciling the two canons. It also carries an AI-era operating section: accountability as the one thing that can never be handed to an agent, comprehension as the binding constraint once code generation is cheap, and encoding the judgment that used to live only in the veterans' heads. Consult for questions about delegation, 1:1s, feedback, meetings, org design, reorgs, decision-making process, who owns the outcome when an agent did the work, founder mode, \"am I micromanaging,\" whether to hire a CPO, or a founder struggling with the IC-to-manager transition. *(Mostly 10+ employees. Below that, delegation is a calendar and first-hires problem rather than an org-design one — go to `time-energy.md` and `hiring.md` instead.)*\n- `references/socratic-technique.md` — The full axiom-to-question translation table, plus how to conduct Socratic mode: question sequencing, Rapoport's Rules steelmanning, deflection handling, the pre-mortem, the calibration and outside-view moves that put explicit odds and a reference class on a founder's forecast (Laplace/Tetlock; Kahneman & Tversky/Flyvbjerg), parking, exit criteria. *(Table and technique, not axioms — read whenever Socratic mode is active.)*\n- `references/meta.md` — Wisdom about wisdom: stage-relevance, founder peer groups, contextuality, how to take advice, Helen Riley's \"rigor kills innovation if applied too early,\" Green's \"early-stage currency is learning, not revenue,\" Megan Smith's scout-before-you-start rule, Skok's communication acid test (can the listener say it back?), Chesky's disbelieve-about-thirty-percent test for whether your judgment is actually updating, and the you-are-the-constant pattern — when the third executive fails the same way, the founder is the one variable that never changed (Sophocles' *Oedipus*)\n\nWhen a question spans multiple domains (and most real founder questions do), read 2–3 files. Don't read more than that unless explicitly asked for a comprehensive scan.\n\n## How to stage-match\n\nA pre-seed founder asking about hiring needs different axioms than a Series C CEO asking the same thing. When you can infer stage from context, prefer axioms tagged for that stage. When you can't, ask one short clarifying question or pick the axioms that hold across stages.\n\nAn axiom carries a `*(Stage: …)*` tag only when it doesn't hold at every stage; an untagged axiom is the corpus's default and is safe to surface to anyone. Treat a tag that doesn't match the founder's stage as a reason to skip the axiom, not to caveat it. But the absence of a tag is that default, not a positive assertion that the axiom was checked at this founder's stage — so when you know the stage and an untagged axiom prescribes a practice that presupposes headcount, a process layer, or a finance function the company doesn't have, cut it down to what this company actually has or leave it out. Adapt or skip is the whole move; it is not a license to hedge the axiom instead.\n\nRough stage map:\n- **Pre-seed / Seed (0–10 people)**: Co-founder dynamics, first hires, product-market fit, raising the first round\n- **Series A (10–50)**: First executive hires, repeatable sales motion, building a real finance function\n- **Series B (50–200)**: Layering management, board sophistication, scaling sales/marketing, real HR\n- **Series C+ (200+)**: Org design, succession, professionalization, IPO-readiness\n\n`Pre-PMF` is a tag value too, and it cuts across the headcount map — match it to whether the company has found product-market fit, not to its round.\n\n## Output style\n\n- Use the **bolded axiom + explanation** format above. The axiom itself should be quotable in one sentence — and if a bolded line isn't quotable on its own as an axiom, it shouldn't be bold.\n- Don't over-hedge. Founders need conviction in advice; weaselly \"it depends\" responses are useless. When something genuinely is contextual, say so plainly and explain the dependency.\n- Don't moralize. These are observed patterns, not commandments.\n- When relevant, name the limit of the axiom. \"Fire fast — except never in anger, never on a Friday\" is more useful than just \"fire fast.\"\n- Avoid stacking unrelated axioms. Better to give three deeply relevant ones than ten generic ones.\n- Benchmarks age; axioms don't. Any figure tagged *[bench YYYY-MM]* in the corpus is a market snapshot, not a law — burn multiples, CAC payback bands, Rule of 40 medians, dilution norms, and multiples all move with the funding environment. Cite the structure with conviction and the number with its vintage: \"the ladder investors use is burn multiple under 1.5x — that's the benchmark as of mid-2026, worth checking current data.\" Never present a dated threshold as timeless.\n- The corpus is sampled from companies that worked; the ones that ran the same play and died didn't write essays. An axiom's recurrence here measures how consistently a pattern shows up among survivors, not how reliably it produces survival. Where one rests on a single company's practice — the Airbnb operating model, the Netflix keeper test, Alphabet X's kill criteria — deliver it with conviction and name the condition it ran on, which is the part that tells this founder whether it transfers: \"don't negotiate your operating model — that's Chesky, and it runs on a founder having more context than anyone they could hire.\" The company's name on its own is decoration; never let \"this worked at Airbnb\" stand in for \"this works.\"\n- Attribute when relevant. Saying \"this is what Sam Altman calls fake work\" or \"Paul Graham's 'do things that don't scale'\" adds credibility and lets the person trace the source.\n\n## What this skill is not for\n\n- Detailed legal advice (term sheets, employment law, IP) — defer to qualified counsel.\n- Specific financial recommendations — defer to a CFO or tax professional. The finance files (`finance-ops.md`, `capital-valuation.md`) give founders the mechanics and the vocabulary to ask better questions; they do not substitute for a CFO, a tax specialist, or securities counsel.\n- Industry-specific operational knowledge (e.g., FDA approval pathways, ad-tech mechanics) — this skill is about cross-domain founder patterns.\n- Reassurance. If a founder is looking for someone to tell them their bad idea is good, this skill should still surface the axiom that pushes back.\n- Bias-free balance. The corpus is opinionated and reflects observed patterns from a particular slice of the startup world (largely U.S., largely venture-backed, largely software). Geographic, sectoral, and structural exceptions exist. Name them when relevant.\n\n## Meta-note on use\n\nThe single most important axiom in the corpus is the meta one: **most advice is contextual; treat it as data, not instruction.** Apply this to the skill itself. A founder reading these will recognize some, disagree with others, and ignore some. That's correct. The value is in the *encounter* with the pattern, not in obedience to it.\n\nFile v1.9.0:evals/README.md\n\n# Evals\n\n`SKILL.md` is prose that steers a model. Prose drifts. This directory holds a small,\nhand-checkable set of scenarios that pin down what the skill is supposed to do:\nwhether it triggers, which reference files it reads, which mode it picks, and\nwhether the output obeys the discipline in `SKILL.md`.\n\nThe directory has two halves. `check_scenarios.py` validates the data file's shape\nand calls no model — stdlib only, no key, runs anywhere Python does.\n`run_scenarios.py` executes the scenarios against a model and needs the `anthropic`\nSDK plus an `ANTHROPIC_API_KEY`. `scenarios.yaml` is the source of truth for both,\nand it is still a file you can read as documentation and run by hand.\n\n## Files\n\n- `scenarios.yaml` — the eval set. 38 scenarios plus five global invariants.\n- `check_scenarios.py` — stdlib-only validator and pretty-printer. It checks the\n  data file's shape, that every referenced path exists, and that `SKILL.md`'s\n  frontmatter description is still inside its 950-character budget (the runtime\n  ceiling is 1024; the gap is deliberate headroom, and the way to get it back is\n  to move enumeration into the routing list, not to raise the budget). It does\n  **not** call a model.\n  It bundles a minimal YAML reader rather than taking a PyYAML dependency; its output\n  on `scenarios.yaml` has been diffed against PyYAML and is identical.\n- `run_scenarios.py` — the executable half. It imports `load` and `validate` from\n  `check_scenarios.py` rather than reparsing the YAML, then drives each `prompt`\n  through a harness that hands the model one skill and two tools, so triggering and\n  file routing are read off the tool calls instead of inferred from the prose.\n  Deterministic checks always run; `--judge` adds a second model call that grades the\n  mode and the prose assertions. Needs the `anthropic` SDK and `ANTHROPIC_API_KEY`\n  unless `--dry-run`.\n- `results/` — committed output from judged runs. See \"Committed results\" below.\n\n## Schema\n\nTop level:\n\n| Key | Meaning |\n|---|---|\n| `version` | Schema version. Currently `1`. |\n| `global_invariants` | Plain-English checks that apply to every scenario where `should_trigger` is true. |\n| `scenarios` | The list below. |\n\nEach scenario:\n\n| Key | Type | Meaning |\n|---|---|---|\n| `id` | string | Stable kebab-case identifier. Referenced in PR discussion. |\n| `prompt` | string | Exactly what the user says. Verbatim — do not paraphrase when running. |\n| `should_trigger` | bool | Whether the founder-wisdom skill should activate at all. |\n| `expected_mode` | `direct` \\| `socratic` \\| `none` | `none` iff `should_trigger` is false. |\n| `expected_files.must_include` | list | Reference files that must appear among those read. Never more than three. |\n| `expected_files.must_not_include` | list | Reference files whose presence is a failure. |\n| `assertions` | list | Plain-English checks on the response. |\n| `rationale` | string | One line naming the `SKILL.md` rule the scenario protects. |\n\n`expected_files` is deliberately tolerant. `SKILL.md` tells the model to read 2–3\nfiles, so pinning an exact set would produce false failures. The pair of lists\nexpresses only what must be there and what must not be — anything else is free.\nWhere a plausible-but-secondary file exists, it is named in an assertion rather\nthan in either list.\n\nThe axiom count is tolerant for the same reason. `SKILL.md` asks for 3–7 and the\nglobal invariant states that rule, but the deterministic check fails only above\nnine. Across three judged runs the bolded-lead-in count centred near six with a\ntail that crossed seven on roughly a third of triggering scenarios, and no\nscenario exceeded seven in every run — so a hard stop at the stated ceiling\nflagged a different handful each time without the skill having changed. The\ninvariant is the target; the check is there to catch a dump, not to police the\nlast two.\n\n## Running against a model\n\n`run_scenarios.py` executes the scenarios instead of printing them. Install the SDK\n(`pip install anthropic`) and set `ANTHROPIC_API_KEY`. `--dry-run` needs neither and\nprints the plan, which is enough to catch a broken runner in CI.\n\n### How the harness models activation\n\nThe model gets the `description:` line from `SKILL.md`'s frontmatter and exactly two\nclient-side tools: `load_skill`, which returns the skill body, and `read_reference`,\nwhich returns one file under `references/`. Nothing else is in context.\n\nSo \"did the skill trigger\" is a `load_skill` call, and \"which reference files were\nread\" is the list of `read_reference` paths — both observed facts about the\ntranscript rather than judgments about the answer, which is the entire reason for\nthe harness. It is an approximation of a real skill runtime, not the runtime itself:\na model that would have behaved differently with a full toolset is not tested here.\n`read_reference` refuses any path that escapes `references/`.\n\n### What runs deterministically\n\nThese need no judge and never disagree with themselves:\n\n| Check | Rule |\n|---|---|\n| Trigger | `load_skill` is called iff `should_trigger` is true. |\n| Stray reads | A non-triggering scenario must read no reference file. |\n| `must_include` | Every named file appears among the reads. |\n| `must_not_include` | No named file appears. |\n| File budget | At most three reference files. |\n| Axiom count | 3–7 bolded axiom lead-ins, skipped in Socratic mode. |\n\nThe file budget is an upper bound only. `SKILL.md` asks for 2–3 files when a question\nspans domains, so a correct single-domain answer reads one — the lower half of that\ninvariant goes to the judge, which can read the prompt and tell the difference.\n\n### What the judge checks\n\n`--judge` adds a second model call per scenario. It grades a fixed claim list: the\nscenario's `expected_mode` restated as a claim, its `assertions`, and — for\ntriggering scenarios — the `global_invariants` the deterministic checks don't\nalready cover. Each claim gets a pass/fail and a one-sentence reason; silence or\nambiguity is a fail. The judge echoes each claim back, and verdicts are matched by\nthat echo rather than by position, so a dropped verdict fails its own claim instead\nof shifting every later one onto the wrong claim.\n\n### Flags\n\n| Flag | Effect |\n|---|---|\n| `--dry-run` | Print the plan. No SDK, no key, no API call. |\n| `--judge` | Add the judged pass. Roughly doubles the calls. |\n| `--id ID` | Run one scenario. Repeatable. |\n| `--limit N` | Run at most the first N. |\n| `--deadline MINUTES` | Stop starting new scenarios and report what ran. |\n| `--json PATH` | Write the full result set, rewritten after every scenario. |\n| `--model` / `--judge-model` | Override either model. |\n| `--threshold F` | Judged pass rate required. Default 0.8. |\n| `-v` | Print every response in full. |\n\nExit codes: `0` clean; `1` a scenario failed on behavior, or the judged score is\nbelow the threshold; `2` bad usage, or `scenarios.yaml` / `SKILL.md` could not be\nused; `3` nothing failed on behavior but the harness errored or ran out of time.\nThe `3` earns its own code — a rate-limit burst is not a skill regression and\nshouldn't be read as one.\n\n### Threshold semantics\n\n`--threshold` is a floor on the judged claims only: the fraction of graded claims\nthat passed, pooled across every scenario that ran, not a per-scenario score and not\na count of scenarios. Deterministic failures are never in that fraction and never\nexcused by it — one behavioral failure exits `1` however high the score. The default\n0.8 is tolerance for judge noise on prose claims, not permission to miss one claim in\nfive.\n\n### The axiom count is a heuristic\n\nNothing here knows what an axiom is. The counter matches the corpus's format instead:\na bolded run opening a line or paragraph, followed by prose on the same line. Bolded\nlabels ending in a colon are dropped, because the corpus writes section labels that\nway itself and counting them would fail a well-formed answer. Against `references/`,\n600 of the 614 line-start bold runs survive and all 14 dropped are labels.\n\nThat makes it a format check wearing a content check's name. A response that surfaces\nfive real axioms in unbolded prose counts zero and fails; one that bolds five\nthrowaway lines counts five and passes. It also can't tell a restated axiom from a\nnew one, or a good axiom from a wrong one. Read a bare axiom-count failure as \"go\nlook at the response\" rather than as a verdict — `-v` prints it.\n\n### Committed results\n\n`evals/results/` holds the output of judged runs, committed. A judged run against\nchanged routing prose then produces a diff you can read in review: which scenarios\nmoved, which claims the judge changed its mind about, which files a scenario started\nreading instead. Drift shows up in the diff rather than in a terminal nobody kept.\n\n## Running by hand\n\nRunning by hand is not what the runner replaced. The runner checks what a machine can\ncheck — triggering, routing, counts, and a judge's read of the claims. Whether the\naxiom it surfaced was the *right* one, whether a Socratic question actually lands,\nwhether the answer would help a founder: those are still yours, and so is checking\nthe skill in the product people use rather than in a harness that approximates it.\n\n1. `python3 evals/check_scenarios.py` — validates the file and prints every scenario.\n2. Install the skill (see the repo README) in a fresh conversation. One scenario\n   per conversation: state carries, and a scenario that follows another will\n   inherit its mode.\n3. Paste the `prompt` verbatim. Nothing else.\n4. Score it:\n   - **Trigger** — did the skill activate? For Claude Code, the tool calls show\n     which `references/*.md` were read. In Claude.ai, ask \"which files did you\n     consult?\" *after* scoring the response, never before.\n   - **Files** — every `must_include` present, no `must_not_include` present.\n   - **Mode** — direct means axioms; Socratic means one question and a stop.\n   - **Assertions** — each one a yes/no.\n   - **Global invariants** — apply to every triggering scenario.\n5. A scenario fails if any single check fails. Record which one; \"it failed\" is\n   not actionable, \"it read `fundraising.md` instead of `capital-valuation.md`\" is.\n\nA full pass by hand is a bit over an hour, and is warranted when `SKILL.md`\nrouting prose changes; a spot check of the affected scenarios is enough for\nanything smaller.\n\n## Running with skill-creator\n\nAnthropic's `skill-creator` skill has eval tooling that runs scenarios against a\nskill and grades the transcript. It reads an `evals/evals.json` whose entries carry\n`id`, `prompt`, `expected_output`, and `expectations`. Translate as follows:\n\n- `prompt` → `prompt`, verbatim.\n- `id` → a short note in `expected_output`; skill-creator's `id` is an integer, so\n  number the entries and keep our kebab-case id as the human label.\n- `should_trigger` → `expected_output` states whether the skill should engage at all.\n  The negatives matter as much as the positives — a skill that fires on everything\n  scores well on positives alone.\n- `assertions` + the matching `global_invariants` → `expectations`, one string each.\n- `expected_files` → an `expectations` entry of the form \"read `references/X.md`;\n  did not read `references/Y.md`.\" Note that skill-creator's own `files` key means\n  *input* files, not the reference files read — do not map onto it.\n\nKeep `scenarios.yaml` as the source of truth and generate the tool's input from it.\nDo not fork the expectations into a second file.\n\n## Adding a scenario\n\n1. It must protect a rule `SKILL.md` actually states. Quote or paraphrase that rule\n   in `rationale`.\n2. If `SKILL.md` is ambiguous on the point, say so in the `rationale` rather than\n   inventing a rule. Four scenarios already do this — `negative-boundary-mixed-regulatory`\n   (mixed in-scope/out-of-scope prompts), `routing-delegation-at-5` (SKILL.md says\n   \"prefer,\" which is softer than the `must_not_include` encoding),\n   `reassurance-still-triggers` (the mode call), and\n   `routing-non-dilutive-revenue-share` (a revenue-based offer is both\n   \"debt-vs-equity\" and \"non-dilutive capital,\" which SKILL.md routes to\n   different files).\n3. Keep `must_include` to the one or two files the routing genuinely requires.\n4. Stay inside the YAML subset the validator's bundled reader handles: two-space\n   indent, `key: value`, `[a, b]` flow lists, block sequences, and `>-` folded\n   scalars. Avoid anchors, inline comments, and multi-line flow collections.\n5. Re-run `check_scenarios.py`.\n\n## Known ambiguities and gaps\n\nThese are recorded so they aren't rediscovered every time:\n\n- **\"Reassurance\" is an output rule, not a trigger rule.** `SKILL.md`'s\n  \"What this skill is not for\" lists reassurance, but the entry itself says the\n  skill \"should still surface the axiom that pushes back.\" `reassurance-still-triggers`\n  encodes the trigger as **true**.\n- **Mixed-scope prompts are unspecified.** A prompt with an in-scope half and an\n  out-of-scope half (`negative-boundary-mixed-regulatory`) has no rule in `SKILL.md`.\n- **\"Prefer\" is soft.** The `management-execution.md` stage note says to prefer\n  `time-energy.md` and `hiring.md` below ~10 people. `routing-delegation-at-5`\n  reads that strictly.\n- **Coverage gap at 200+.** `hiring.md`'s stage tags are all Seed–Series A, so\n  `output-stage-match-late` tests stage-awareness against thin material.\n- **`mode-direct-factual` is unverified since its corpus fix.** The scenario asserts\n  the option-pool percentage carries a vintage tag, which is what `SKILL.md`'s\n  \"Output style\" requires. It used to fail on the corpus rather than on `SKILL.md`:\n  the 10–15% pool-refresh figure in `capital-valuation.md` carried no\n  `*[bench YYYY-MM]*` tag, so a model reading it faithfully had nothing to cite. The\n  figure is now tagged, which removes the known cause — it does not establish a pass,\n  because no run has been made since. Re-run the scenario before treating it as green.\n\nFile v1.9.0:README.md\n\n# Founder Wisdom\n\nA [Claude Skill](https://www.anthropic.com/news/skills) that surfaces hard-won axioms from experienced startup founders and operators — the pattern-matched wisdom that first-timers usually learn by running into walls.\n\n## What it does\n\nFounders face decisions where the right answer isn't obvious from first principles — it's obvious from pattern. This skill gives Claude access to a curated corpus of those patterns, organized by domain, with the context that makes each one useful.\n\nWhen you ask Claude a founder-stage question — hiring, fundraising, product, sales, finance, co-founders, governance, crisis, or any of the surrounding territory — Claude consults this corpus and responds with the relevant axioms rather than working from generic priors.\n\nThe skill operates in two modes:\n\n- **Direct mode** (default): Surfaces relevant axioms with reasoning and qualifiers. Useful when you want a quick read on a decision.\n- **Socratic mode**: Asks the question the axiom answers, rather than handing you the answer. Useful for working through ambivalence, for coaching/mentoring conversations, and for processing live decisions.\n\n## Domains covered\n\n- Hiring, firing, comp, equity grants, the keeper test, and the AI-era talent shift toward systems thinkers\n- Fundraising, terms, runway, investor relations\n- Product, product-market fit, pivots, and where startup ideas come from — Paul Graham on noticing rather than inventing, and on schlep blindness\n- Sales, pricing, GTM, the founder-as-seller, what a visionary customer is actually buying, and selling in the AI era — building the motion like a product, automating in order of workflow legibility\n- Finance, cash, burn, CFO timing, ramen profitability, AI-pilot revenue and ERR, the operating metrics that matter (cohorts, NRR, Rule of 40, CAC payback, gross margin), and how to build the two models the axioms keep pointing at — the 13-week cash flow forecast and the Default Alive calculation\n- Capital and valuation — cost of capital, dilution math, venture debt, ARR multiples, the liquidation waterfall, 409A, down rounds, secondaries, plus the procedures: computing your own waterfall off the charter, and running a down round\n- Bootstrapping and the non-venture path — whether to raise at all, self-financeable growth, overtrading, annual prepay, revenue-based financing, SBIR grants, profit share and phantom equity, the plateau\n- Exits and M&A — running a sale process, bankers, LOIs and exclusivity, diligence, earn-outs, escrow, retention packages, acqui-hires, headline price vs. actual payout\n- Co-founder dynamics, splits, vesting, titles, and Noam Wasserman's rich-versus-king trade\n- Boards and governance — board design, running the board like a team you lead, and why success is what gets founder-CEOs replaced\n- Founder time, energy, sustainability, and where founder attention goes — the maker's schedule, the one top idea in your mind\n- Customers and market dynamics, and Geoffrey Moore's chasm — why early adopters can't reference you to the mainstream\n- Strategy, moats, and defensibility — Helmer's 7 Powers (including the math), Rumelt's kernel, Thiel, Greenwald, Christensen, Porter\n- Management and execution — the IC-to-manager transition, delegation, 1:1s, meetings, org design, reorgs, the AI-era operating axioms, and the founder-mode counter-canon (Brian Chesky, Paul Graham)\n- Crisis, layoffs, resilience, not dying as most of the strategy, and how to wind down when the answer is to stop — creditor priority, payroll withholding as trust money, dissolution vs. an assignment for the benefit of creditors vs. Chapter 7, the reserve that buys an orderly shutdown, and who you owe in what order\n- Culture, vision, the cultural DNA of the first 20 employees, and talent density as the input every other freedom depends on\n- Startup mechanics: incorporation, founder stock, IP assignment, vesting, and the founder exposures outside the corporate veil — personal guarantees, veil-piercing, trust-fund payroll taxes, claims-made D&O and tail coverage, sponsored immigration status, director duties near insolvency\n- The YC canon, in two files — the essays (Paul Graham, Sam Altman, Jessica Livingston, Geoff Ralston, Paul Buchheit, Aaron Harris, Michael Seibel on fundraising) and the product/PMF material (Michael Seibel, Gustaf Alströmer, Eric Migicovsky, Dalton Caldwell, Peter Reinhardt)\n- The Harvard Innovation Labs canon — talks hosted at the i-lab from Michael Skok's *Startup Secrets* series, Underscore VC partners (Lily Lyman, Chris Gardner), Alphabet X (Helen Riley), Megan Smith (3rd U.S. CTO), Phil Green, and Rebekah Emanuel\n- Meta-wisdom: how to take advice, contextuality, founder peer groups\n\n## Installation\n\nSkills work in Claude.ai, Claude Code, and the Claude API. For other LLMs, see [Using with other LLMs](#using-with-other-llms) below.\n\n- **Claude Code**: clone this repository (or copy `SKILL.md` and `references/`) into `~/.claude/skills/founder-wisdom/` for personal use, or `.claude/skills/founder-wisdom/` inside a project.\n- **Claude.ai and the Claude API**: zip the repository contents (with `SKILL.md` at the root of the zip) and upload it through your client's skill settings.\n\nFor more on Claude Skills, see [Anthropic's skills documentation](https://docs.claude.com).\n\n## Using with other LLMs\n\nThe corpus itself is plain provider-neutral markdown; only the skill packaging (frontmatter triggering, file-based progressive disclosure) is Claude-specific. The `dist/` directory carries generated artifacts for everything else:\n\n- **`dist/founder-wisdom-full.md`** — the entire skill (routing guidance plus all twenty-one reference files) as one self-contained document, with file cross-links rewritten as section references. At roughly 75–80k tokens it fits in current large context windows: paste it into a Gemini Gem's instructions, a ChatGPT Project, or any long-context model's system prompt.\n- **`dist/system-prompt.md`** — a short system prompt for platforms where the corpus lives in a retrieval store rather than in context (e.g. a ChatGPT Custom GPT with the reference files uploaded as knowledge). It carries the trigger conditions, the two modes, the section roster with routing notes, the stage-matching rules, and the output-style rules, plus an instruction to retrieve whole topic sections by name to compensate for chunk-based retrieval.\n\nBoth files are generated — never edit them by hand. The sources (`SKILL.md` and `references/`) remain the single source of truth; after changing them, regenerate with:\n\n```bash\npython3 scripts/build_bundle.py\n```\n\nThe script is deterministic and fails loudly if the sources drift out of sync with its rewrites (e.g. a reference file missing from SKILL.md's routing list). To verify without writing anything — a build in memory, compared against the committed files, exiting non-zero when they differ:\n\n```bash\npython3 scripts/build_bundle.py --check\n```\n\nThat is the command CI runs on every pull request. CI runs it against the committed tree while a local run reads your working copy, so commit the regenerated `dist/` alongside the sources it came from — a `--check` that passes on artifacts you rebuilt but never staged will still fail in CI.\n\n## Use cases\n\nThe skill is designed to be useful for several distinct audiences:\n\n- **Founders** wrestling with a specific decision or just trying to pattern-match where they are\n- **Operators** (CEOs, COOs, VPs) who didn't found the company but face the same operating questions\n- **Investors and advisors** who want to ground their guidance in the canonical wisdom\n- **Executive coaches, mentors, and peer-advisors** working with founder-stage clients (Socratic mode is particularly designed for this)\n- **Aspiring founders** learning the landscape before they start\n\n## Philosophy\n\nThe corpus is **opinionated and pattern-matched, not neutral or comprehensive**. Every axiom in it has counter-examples. The value is in the encounter with the pattern, not in obedience to it.\n\nThe corpus is also **biased toward observed patterns from a particular slice of the startup world**: largely U.S., largely venture-backed, largely software. Geographic, sectoral, and structural exceptions exist and should be named when relevant.\n\nThe corpus is **sampled on outcome**: almost everything in it comes from founders and companies that worked. An axiom earns its place by how consistently the pattern shows up among survivors, not by how reliably it produces survival — the companies that ran the same play and died mostly didn't write essays. That is a different bias from the slice above and compounds with it, so where an axiom rests on a single company's practice the skill names the condition that practice ran on rather than presenting the result as a general law.\n\nThe skill is **not a substitute for qualified legal, financial, or specialized professional advice**. It's a thinking partner that knows the canonical wisdom, not an oracle.\n\n## Attribution\n\nThe YC canon section attributes axioms to their sources: Paul Graham, Sam Altman, Jessica Livingston, Michael Seibel, Geoff Ralston, Paul Buchheit, and Aaron Harris, drawn from essays linked in [YC's Essential Startup Advice](https://www.ycombinator.com/blog/ycs-essential-startup-advice). The original essays remain the property of their authors. This skill paraphrases and synthesizes — it does not reproduce them.\n\nPaul Graham's essays also supply material in the domain files outside the two canon files: \"The Equity Equation,\" \"Maker's Schedule, Manager's Schedule,\" \"The Top Idea in Your Mind,\" \"How Not to Die,\" \"Ramen Profitable,\" \"Default Alive or Default Dead?,\" \"Startup = Growth,\" \"How to Get Startup Ideas,\" \"Schlep Blindness,\" and \"Founder Mode.\" Each is cited inline where it is used; all are published at [paulgraham.com](https://paulgraham.com/articles.html).\n\nThe Harvard Innovation Labs canon is drawn from talks given at the i-lab — including Michael Skok's *Startup Secrets* workshops, Lily Lyman and Chris Gardner of Underscore VC, Helen Riley (CFO/COO of Alphabet X), Megan Smith (3rd U.S. CTO), Phil Green, and Rebekah Emanuel. Much of that material is published on the [Harvard Innovation Labs YouTube channel](https://www.youtube.com/@harvardilab), but not all of it is: where an axiom rests on a talk with no published recording, the reference file says so inline rather than letting the attribution read as a citation. The original talks remain the property of the speakers and Harvard Innovation Labs. This skill paraphrases and synthesizes — it does not reproduce them.\n\nAxioms attributed to Brian Chesky, Jeanne DeWitt Grosser, and Elizabeth Stone are drawn from three episodes of Lenny's Podcast: Brian Chesky (Airbnb), \"Brian Chesky's new playbook\"; Jeanne DeWitt Grosser (Vercel; previously Stripe and Google), \"What world-class GTM looks like in 2026\"; and Elizabeth Stone (CPTO, Netflix), \"Why Netflix is betting on systems thinkers - not specialists - in the AI era.\" The episodes remain the property of the speakers and the podcast. This skill paraphrases and synthesizes — it does not reproduce them.\n\nThe strategy and management sections draw on frameworks from published books: Hamilton Helmer's *7 Powers*, Richard Rumelt's *Good Strategy/Bad Strategy*, Peter Thiel's *Zero to One*, Bruce Greenwald's *Competition Demystified*, Clayton Christensen's *The Innovator's Dilemma*, and Michael Porter's work (via Joan Magretta) in `references/strategy-moats.md`; and Andy Grove's *High Output Management*, Ben Horowitz's *The Hard Thing About Hard Things*, Matt Mochary's *The Great CEO Within*, and Elad Gil's *High Growth Handbook* in `references/management-execution.md`. These books remain the property of their respective authors and publishers. This skill paraphrases and synthesizes their frameworks into a different format — it does not reproduce their text. Readers who want the full argument should consult the original books directly.\n\nThe co-founder, governance, and market sections draw on three more books: Noam Wasserman's *The Founder's Dilemmas* in `references/cofounders-equity.md` and `references/governance.md`; Geoffrey Moore's *Crossing the Chasm* in `references/customers-market.md`, `references/product.md`, and `references/sales-gtm.md`; and Brad Feld, Matt Blumberg, and Mahendra Ramsinghani's *Startup Boards* in `references/governance.md`. Three shorter sources sit alongside them — Thomas Hellmann and Noam Wasserman's paper \"The First Deal,\" Fred Wilson's board posts on [AVC](https://avc.com), and the one-pager \"Expectations for Outside Board Members,\" written by a portfolio CEO and reprinted on Brad Feld's blog, which the corpus keeps distinct from *Startup Boards*. These remain the property of their authors and publishers. The same rule applies: this skill paraphrases and synthesizes their frameworks — it does not reproduce their text.\n\nBeyond the startup canon, a newer layer of the corpus reaches to older sources for its risk, incentive, and forecasting axioms — each cited inline where it is used, and introduced to the corpus for the first time here. Shakespeare's *King Lear*, who banishes the daughter who tells him the truth, stands behind the bad-news axiom in `references/culture.md`; Sophocles' *Oedipus the King*, who hunts the outsider whose corruption turns out to be himself, behind the you-are-the-constant pattern in `references/meta.md`; Charlie Munger's incentive-caused bias behind the advisor-incentive scan in `references/meta.md`, and his Persian-Messenger discussion — with *King Lear* — behind the bad-news axiom; Peter Thiel's power law from *Zero to One*, here in its founder-side sense, the fund's payoff distribution seen from the one path the founder travels, in `references/capital-valuation.md`, distinct from the *Zero to One* strategy material cited above; Ole Peters on ergodicity, the Kelly criterion, and Nassim Taleb on risk of ruin behind the survive-the-bet axiom in `references/finance-ops.md`; Jay Forrester and Donella Meadows on feedback delay and overshoot in system dynamics behind the signal-lag whipsaw in `references/finance-ops.md`; Daniel Kahneman and Amos Tversky's inside-versus-outside view with Bent Flyvbjerg's reference-class forecasting behind the reference-class downside in `references/finance-ops.md` and the outside-view move in `references/socratic-technique.md`; and Pierre-Simon Laplace's rule of succession with Philip Tetlock's forecasting research behind the calibration move in `references/socratic-technique.md`. These plays, papers, and books remain the property of their authors and publishers; as everywhere else in this corpus, the skill paraphrases and synthesizes — it does not reproduce their text.\n\nThe non-canon axioms in the corpus reflect widely-circulated startup wisdom that has no single attributable source. Where a specific person originated an axiom, attribution is given inline.\n\n## Structure\n\n```\nfounder-wisdom/\n├── SKILL.md                          # The main routing logic for Claude\n├── evals/                            # Behavioral scenarios that pin down SKILL.md\n│   ├── scenarios.yaml\n│   ├── check_scenarios.py\n│   └── README.md\n├── scripts/\n│   └── build_bundle.py               # Generates dist/ from SKILL.md + references/\n├── dist/                             # Generated artifacts for non-Claude LLMs (do not edit)\n│   ├── founder-wisdom-full.md\n│   └── system-prompt.md\n└── references/\n    ├── hiring.md\n    ├── fundraising.md\n    ├── product.md\n    ├── sales-gtm.md\n    ├── finance-ops.md\n    ├── capital-valuation.md\n    ├── bootstrapping.md\n    ├── exits-ma.md\n    ├── cofounders-equity.md\n    ├── governance.md\n    ├── time-energy.md\n    ├── customers-market.md\n    ├── crisis-resilience.md\n    ├── culture.md\n    ├── startup-mechanics.md\n    ├── strategy-moats.md\n    ├── management-execution.md\n    ├── socratic-technique.md\n    ├── yc-canon.md\n    ├── yc-canon-product.md\n    └── meta.md\n```\n\nThe reference files are designed to be readable on their own as well as consumed by Claude. If you want to browse the corpus directly, start with `references/meta.md` and `references/yc-canon.md`.\n\n`references/yc-canon.md` and `references/yc-canon-product.md` are companion halves of the same canon: the first carries the essay-derived axioms and YC's pocket guide, the second the product and product-market-fit material.\n\n`evals/` holds a small set of hand-checkable scenarios covering triggering, routing, mode selection, and output discipline. It is data plus instructions, not a test harness — see `evals/README.md`.\n\n## Contributing\n\nContributions are welcome. See [CONTRIBUTING.md](CONTRIBUTING.md) for what makes a good axiom, what kinds of additions are likely to be accepted, and what kinds are likely to be rejected.\n\nA few general principles:\n\n- **The corpus is opinionated by design.** \"Add both sides\" is usually the wrong instinct. If you have a counter-axiom that's genuinely true, propose it — but be willing to defend it as observed pattern, not just as balance.\n- **Axioms must be quotable in one sentence.** If the axiom needs three sentences to state, it's not an axiom — it's an essay.\n- **Attribution matters.** If you can trace an axiom to a specific person, name them.\n- **Pattern, not just opinion.** The bar is \"this has been observed across many companies,\" not \"this seems right to me.\"\n\n## License\n\nThis skill is released under the MIT License. See [LICENSE](LICENSE) for details.\n\nThe axioms themselves represent widely-circulated startup wisdom. The synthesis, organization, and skill machinery are MIT-licensed. The original YC essays remain the property of their authors; consult them directly via the sources named in `references/yc-canon.md`, `references/yc-canon-product.md`, and in the [YC Startup Library](https://www.ycombinator.com/library).\n\n## Acknowledgments\n\nThis skill stands on the shoulders of the founders, operators, and investors who wrote down what they learned — particularly the Y Combinator partners whose essays form the YC canon section, and the speakers at Harvard Innovation Labs whose recorded talks form the i-lab canon section. It also reflects the broader community of operators who have made startup wisdom a generally accessible body of knowledge over the past two decades.\n\nFile v1.9.0:_meta.json\n\n{\n  \"ownerId\": \"kn7bwxjecdr5xfbq5tf88jsya583f0sv\",\n  \"slug\": \"founder-wisdom\",\n  \"version\": \"1.9.0\",\n  \"publishedAt\": 1786553282130\n}\n\nFile v1.9.0:references/bootstrapping.md\n\n# Bootstrapping & the Non-Venture Path\n\nMost of this corpus is calibrated to companies that raise institutional money, because that is where most of the observed pattern comes from. This file is the counterpart to that material rather than a rebuttal of it: the venture axioms hold for venture companies, and what breaks here is the subset that quietly assumes a round standing behind them. What changes is the binding constraint — growth is rate-limited by margin instead of by a fundraising calendar, nobody outside the company grades the number, and the characteristic death is a decade-long plateau rather than a zero-cash date. `capital-valuation.md` prices the instruments; this file covers what the absence of a round does to everything else.\n\nThis file is not financial, tax, or legal advice. The math is here so founders can ask their CFO and counsel better questions, not so they can skip them.\n\n> **Benchmarks are calibration, not truth.** Figures marked *[bench 2026-07]* — the revenue ceiling that separates a good company from a viable venture asset, annual-prepay discounts, revenue-based-financing caps and remittance rates, SBIR award sizes, profit-share pool sizing, small-business exit multiples — reflect market conditions as of July 2026 and move with the rate environment and the funding market. The mechanics are durable; the numbers are not. Check current data before quoting a threshold, and name the vintage.\n\n## Core axioms\n\n**You can raise later; you cannot un-raise.** The first institutional dollar deletes \"good enough\" as an available outcome: a $250M fund holding 10% at exit needs a $2.5B result to return itself once, so the $80M acquisition that clears your mortgage rounds to zero on the board's math (`capital-valuation.md`). Name the fund's size against the honest ceiling of the business before the first meeting; if it can't cover, you are volunteering to be a write-off. The limit is matching the fund to the outcome: a $30M pre-seed fund returns itself on a $300M exit. Later has arrived when you can name a payback period you cannot fund from margin; it hasn't when you can't say what the money buys.\n\n**The growth plan in the deck is one you'd never run with your own money.** Three shapes take venture money badly. Services-inflected businesses get funded on a software multiple and spend three years starving the delivery motion that was working. Slow compounders with long sales cycles or regulated buyers are excellent on a ten-year clock and unfundable on a fund's. The real-but-small market is the most honest: a business that tops out at $40M *[bench 2026-07]* is a very good company and a bad venture asset, and the deck that clears diligence is the one with an inflated TAM. The limit is services-as-wedge, where the mismatch is deliberate and time-boxed.\n\n**Margin-funded growth is capped by the gap between paying for a customer and being paid.** The venture path pre-funds that gap with a round; here it is funded by last quarter's profit, which makes the ceiling arithmetic — operating cash margin divided by the cash each new dollar of revenue consumes, compounded once per cash cycle (Neil Churchill and John Mullins' self-financeable growth rate). A company keeping ten cents on the dollar and laying out fifty cents four months before collecting tops out near 20% a cycle — call it 70% a year, a third of Livingston's 10%-a-month floor (`yc-canon.md`). It inverts when the customer pays first, which is why `capital-valuation.md`'s working-capital levers are the growth strategy here, not housekeeping.\n\n**A growth rate nobody chooses defaults to whatever inbound showed up.** The venture benchmarks in `yc-canon.md` assume a path where a slow quarter is a fundraising event; strip the round out and the only party grading the number is the bank balance. The first failure mode is importing the 10%-a-month floor as a moral standard and spending years of retained earnings chasing a rate no one will reward. The second is quieter and more common: never setting a rate at all, so the company discovers its strategy a year after it executed it. Name the rate, price it in cash months against the ceiling above, and re-decide it out loud every year.\n\n**You can go broke growing.** Every new customer consumes cash before it returns cash, so a company adding customers faster than its cycle converts them digs a hole that deepens with each win — the trade-finance name is overtrading, and it kills more profitable companies than losses do. The P&L shows a profit the whole way down; the tell is three consecutive months of revenue up, margin flat, and cash down, which is what `finance-ops.md`'s 13-week cash flow forecast exists to catch. This is a balance-sheet problem with balance-sheet fixes (`finance-ops.md`), and the founder who cuts growth when the defect is net-45 invoicing trades a solvable problem for a permanent one.\n\n**A price increase and a new logo are not the same dollar.** Rank revenue sources by the cash each consumes before it returns and the ends carry the argument: a price increase lands next billing cycle at zero working capital; paid acquisition is the most cash-hungry and reprices without your consent (`finance-ops.md` on Gurley). A venture-funded competitor can ignore the ordering because the round buys the expensive end; you cannot, so pricing and retention come first and the sales team gets hired out of the cash they produce. The limit is that the cheap end runs out: pricing power is bounded by the size of the base, so this is a sequence, not a permanent preference.\n\n**Annual prepay is the only interest rate most bootstrapped founders ever negotiate.** A 10–20% discount *[bench 2026-07]* is not a 10–20% annual rate; `capital-valuation.md` carries the annualization and what the discount actually prices out at. The structural trap is the one the discount hides: a company funding this year's operations out of next year's prepay has to grow bookings every single year just to stay level, and the year it doesn't feels like a collapse rather than a flat quarter.\n\n**A revenue-share cap is not a price; the price is the APR.** Revenue-based financing is quoted as a flat multiple — $500K against a 1.2x cap, remitted as 6–10% of monthly revenue *[bench 2026-07]* — and a multiple is silent on time, where the whole cost lives: repay 1.2x over twenty-four months and you paid 18% a year; over twelve because you grew, closer to 40%. Growth accelerates the remittance, shortens the term, and raises the rate — the instrument charges most precisely when it works. It is honestly priced against an acquisition loop whose payback beats the repayment window (`capital-valuation.md`), predatory in the terms the multiple hides — no prepayment discount, daily remittance, a blanket lien, a personal guarantee — and worst when it funds burn. *(Stage: $500K ARR and up — no lender underwrites a revenue history you don't have.)*\n\n**Grants are non-dilutive in equity and expensive in direction.** SBIR is the instrument that matters — $300K for Phase I, up to $2M for Phase II *[bench 2026-07]*, on a six-to-nine-month cycle you cannot accelerate — and at the defense agencies it buys a government customer and a contracting vehicle worth two years of selling. The trap is the treadmill: a company that gets good at winning awards writes its roadmap to solicitations instead of to customers, and the tell is year four with $6M of awards and no commercial revenue. Eligibility turns on majority ownership by U.S. individuals or eligible small businesses; only agencies that opted into a separate authority accept majority VC, PE, or hedge-fund ownership.\n\n**A personal guarantee takes shutting down off the table.** Savings put in is a clean trade: you bought equity at a price you set, bounded by the check. A guarantee on the lease, a credit line against the house, or payroll on personal cards converts a company failure into a personal one, and a founder who has guaranteed the debt cannot kill the company, because killing it realizes the loss. The longest-running zombie companies are reliably the personally-guaranteed ones, kept alive by a founder servicing an obligation rather than running a business. Write down the number you can afford to lose before you start and treat it as a ruin constraint rather than a budget (`finance-ops.md`). The guarantee's own mechanics — what's negotiable in it, and the rest of what sits outside the corporate veil — are in `startup-mechanics.md`.\n\n**With no exit on the roadmap, an option grant is a story, not compensation.** An option converts into money at a liquidity event and nowhere else, so at a company that intends to stay private and profitable its honest expected value is zero. The discovery moment is predictable: a good employee resigns and learns they have 90 days to write a real check for stock no one will buy, plus tax on a paper spread. Grant them for the symbolism if you like, but price them at zero in your own head, say in the offer conversation that there is no exit planned, and pay the difference in cash. What the alternative costs isn't a lawsuit; it's the next three years of compensation conversations.\n\n**Pay at the top of the cash band; share profit on a published formula.** Both are ordinary income and hit the P&L, not the cap table; profit share funds itself, paying only in the years there is something to pay from. Size it as a fixed share of pre-tax profit — a 10–20% pool is a common design point *[bench 2026-07]* — publish the formula and pay on a fixed date; a discretionary bonus decays into noise within two cycles. 37signals pays top-of-San-Francisco-market for every role regardless of location and issues no options at all (Jason Fried and David Heinemeier Hansson, *It Doesn't Have to Be Crazy at Work*). A funded competitor structurally cannot match it: its cash belongs to a burn plan its board resets every eighteen months.\n\n**Phantom equity is the honest version of the option grant.** What makes a plan real is the pair: a valuation method written down and applied every year whether the number went up or down, and a payout trigger the company can fund without a financing. Missing either, it is the same lottery ticket in a new wrapper. The instrument is thin — notional units that pay cash on named triggers, no stock issued, no minority shareholder to buy out later. Ordinary-income treatment is the general case, and deferred-compensation rules govern the design, so the plan document is counsel's call. *(Stage: past 15 employees, where an annual valuation earns its cost.)*\n\n**Once founder pay sets the household budget, every reinvestment is a pay cut.** Past the point where founder pay clears what the market would pay a replacement, the excess is retained earnings you have already spent. It never appears as a line item — nobody writes down \"we didn't hire the second salesperson because we take $400K out\" — so the ceiling gets set by a decision no one records. The fix is bookkeeping: split founder pay on paper into a market salary and an owner distribution set once a year. An acquirer runs the same normalization in diligence: what costs you at exit is never the money you took out, which gets added back, but the reinvestment you skipped. *(Stage: once founder pay clears market replacement — past $1–2M revenue.)*\n\n**The bootstrapped failure mode is not zero cash; it is a comfortable plateau.** Companies on this path rarely die; they plateau between covering the founders comfortably and a few million in revenue, then hold that line for a decade. `finance-ops.md`'s ramen axiom gestures at this; here it is the central risk, because nothing is structurally left to force the question. The cost lands at the end: a no-growth, owner-dependent business prices as a small business — a low-single-digit multiple of owner earnings, not of revenue *[bench 2026-07]* — and often has no buyer at all. The limit: a deliberate lifestyle business is the same company on every metric and a legitimate outcome — the difference is whether the founder decided, and what has no defenders is the undecided state. *(Stage: past founder-supporting profitability — $1M revenue or five people.)*\n\n**Keep the option to raise alive at maintenance cost.** That means a Delaware C-corp, a cap table with no handshakes on it, and books a stranger could audit. The LLC that saves real money in year two costs legal fees to convert and leaves every dollar of appreciation earned before the conversion outside the QSBS exclusion (`startup-mechanics.md`) — incorporate as though you might raise even while you're certain you won't. The maintenance runs a few thousand dollars a year; reconstructing it inside live diligence runs three to six months at the moment your leverage is most expensive. The limit: phantom-equity and profit-share plans are contractual claims on cash flow a preferred investor will want unwound before the wire, so write them terminable on a financing.\n\n## Common founder mistakes\n\n- Pitching a business with an honest $40M ceiling *[bench 2026-07]* to a fund whose math requires a billion-dollar outcome, then spending three years being held to the TAM the deck had to invent.\n- Reading a revenue-based financing offer as a 1.2x multiple rather than converting it to an APR, and discovering the rate went up because the growth arrived.\n- Reading an annual-prepay discount as an annual interest rate. It buys five and a half months of acceleration, not twelve, which prices it at two to three times the number on the discount line.\n- Using revenue-based financing to fund burn instead of an acquisition loop with a measured payback — 30%-plus money buying time with no mechanism to convert it.\n- Personally guaranteeing the lease or the credit line, and then being unable to shut down because closing the company is the act that realizes the loss.\n- Managing off a monthly income statement that shows a profit while the bank balance falls. That is the shape overtrading always has, and the P&L will never name it.\n- Quoting option value in an offer conversation at a company with no exit on the roadmap. The first departure teaches the whole team what the grant was worth.\n- Running a discretionary bonus in place of a published profit-share formula. It decays into noise within two cycles and stops being read as compensation at all.\n- Writing the roadmap to grant solicitations instead of to customers, and arriving at year four with a strong award record and no commercial revenue.\n- Incorporating as an LLC to avoid the second layer of tax, then paying to convert and discovering that the appreciation earned before the conversion sits outside the QSBS exclusion.\n\n## What to read\n\n- Neil Churchill and John Mullins, \"[How Fast Can Your Company Afford to Grow?](https://hbr.org/2001/05/how-fast-can-your-company-afford-to-grow)\" *HBR*, May 2001 — the self-financeable growth rate, and the rigorous version of the ceiling above.\n- Jason Fried and David Heinemeier Hansson, *It Doesn't Have to Be Crazy at Work* — growth rate, comp policy, and profit share stated as deliberate choices rather than as constraints.\n- Rob Walling, *Start Small, Stay Small* — the operating playbook for the deliberately small market, from the founder who co-founded MicroConf and TinySeed around it.\n- [SBIR's venture-capital ownership authority](https://www.sbir.gov/vc-ownership-authority) — the exception to the majority-U.S.-individual ownership requirement: which agencies opted in, and the single-firm limit that applies when they do.\n- Paul Graham, \"[Ramen Profitable](https://www.paulgraham.com/ramenprofitable.html)\" — the milestone this entire path is built on top of, and the one it most easily gets comfortable at.\n\nFile v1.9.0:references/capital-valuation.md\n\n# Capital & Valuation\n\nEverything about the *price* of money and the *value* of the company. `fundraising.md` covers running the raise; `finance-ops.md` covers spending the cash once it's in the bank. This file covers the layer in between: what capital actually costs, how the company gets valued, and what the founder ends up owning.\n\nThis file is not financial, tax, or legal advice. The math is here so founders can ask their CFO and counsel better questions, not so they can skip them.\n\n> **Benchmarks are calibration, not truth.** Figures marked *[bench 2026-07]* — implied cost of equity, payback bands, venture debt sizing, prepay discounts, round-by-round dilution, Rule of 40 medians, the 409A-to-preferred gap, the preference-multiple norm, the priced-round timeline — reflect market conditions as of July 2026 and move with the rate environment. The mechanics are durable; the numbers are not. Check current data before quoting a threshold, and name the vintage.\n\n## Core axioms\n\n**Equity is the most expensive capital you will ever raise.** Founders treat equity as free because no interest payment shows up in the P&L. It isn't. The implied cost of venture equity is the return the fund needs to make on you — a seed fund underwriting 10x on its winners is charging you something on the order of 40–60% a year, compounding, forever. Debt at 12% is dramatically cheaper money. The reason founders take equity anyway is that it's *patient* and *loss-absorbing*, not that it's cheap. Say the sentence out loud before every round: \"I am about to take out the most expensive loan available to me, and it never gets repaid.\"\n\n**Your investor is paid on the power law; you are paid on one path through it.** (Peter Thiel, *Zero to One*.) A fund's return is dominated by its handful of biggest winners, so it rationally prefers a 20% shot at a $1B outcome to your near-certain $60M one — same bet, opposite optimum, because the fund holds a portfolio of these and you live only this one. Read persistent board pressure to raise more, hire ahead of revenue, and swing bigger as portfolio math, not as a read on your odds of survival. Weight any piece of risk advice by how much of your downside the person giving it actually carries — taking secondary to lower your own risk aversion (below) is the standard, partial way that gap gets closed. Portfolio math is why they push for variance; the limit that overrides it regardless — never a bet you can't survive losing — is in `finance-ops.md`.\n\n**Every equity trade has a break-even, and most founders never compute it.** (Paul Graham, \"The Equity Equation.\") Giving up n% pays for itself only if what you get back raises the company's average outcome enough that the smaller slice beats the whole you held before — the implied-cost argument above, applied one deal at a time rather than one round at a time. See `cofounders-equity.md` for the full treatment.\n\n**Use payback period, not NPV, as your internal hurdle.** No seed-stage founder should be running discounted cash flow models on a hiring decision. But every founder should be able to answer \"how many months until this dollar comes back?\" for any material spend — a rep, a marketing channel, a new market. The rough ladder for venture-stage software *[bench 2026-07]*: under 12 months is excellent, 12–18 is standard, 18–24 needs a specific thesis, over 24 months means you're funding it with equity at 40%+ implied cost and should say so out loud. Payback is the founder-legible stand-in for cost of capital, and it has the useful property of being honest about *time*, which is the thing that actually kills startups.\n\n**Your working capital is a free credit line, and most founders never draw on it.** Invoicing on signature instead of on delivery, net-15 instead of net-45, collecting on time — these move real cash into the company at a cost of capital of *zero*, and they compound. Annual prepay instead of monthly billing moves more cash than all three, and it is the only one that costs anything. A $4M ARR company that shifts from monthly to annual-prepay on half its base pulls roughly $1M of cash forward and buys itself months of runway without a single conversation with an investor. The price is a discount (10–20% is market for annual prepay *[bench 2026-07]*) and some deal friction. Annualize that discount before you decide it's cheap: the cash it accelerates was arriving an average of five and a half months out, not twelve, so the discount buys five and a half months of acceleration rather than a year and prices out near 25–60% annualized *[bench 2026-07]* — comparable to revenue-based financing (priced as an APR in `bootstrapping.md`) at the low end, dearer at the high end, and at the top of the band it costs more than the equity it was supposed to beat. Take it anyway when cash is the constraint — it is the one cost of capital you set rather than accept, and the only one you can reprice at the next renewal. What it does to a company with no round behind it — this year's operations funded out of next year's bookings, so bookings have to grow every year just to stay level — is in `bootstrapping.md`.\n\n**Deferred revenue is a loan from your customers — cheap, but it is a liability.** The cash from an annual prepay is not income; it's an obligation to deliver twelve months of service. It funds you interest-free, which is excellent. It also means a churned customer in month three is a real refund conversation and a real hole. Founders who spend deferred revenue as though it were earned revenue are borrowing against a promise they may not be able to keep.\n\n**Venture debt is bought with covenants, not with interest.** The headline rate on venture debt looks cheap next to equity, and the dilution is genuinely lower. That's not where the cost is. The cost is in the covenants, the material-adverse-change clause, and the fact that the lender's collateral is your company. Venture debt is safe when you are certain of the next equity round and want to extend runway into a stronger milestone. It is lethal when it's a substitute for a round you can't raise — the covenant trips exactly when you're weakest, and the lender's interests and yours diverge precisely at the worst moment. Rule of thumb: only take venture debt you could survive having called. *(Stage: Series A+.)*\n\n**Size debt against a milestone, not against a gap.** Standard venture debt is roughly 25–35% of the last equity round *[bench 2026-07]*. The correct question isn't \"how much can I get?\" but \"what milestone does this money buy, and is that milestone worth more than the covenant risk?\" Debt that buys you from $2M to $4M ARR before a Series A is a good trade. Debt that buys you four more months of the same is a slow-motion assignment of the company to the lender.\n\n**Dilution compounds, and the seed round is the one you're most likely to over-optimize.** Founders fixate on the dilution of the round in front of them. The number that matters is the product of every round *[bench 2026-07]*: 20% at seed, 20% at A, 18% at B, 15% at C, plus an option pool refresh at each — a founding team that started with 100% is in the mid-40s by Series C on price dilution alone, and in the low 30s once the pool refreshes are counted. Two implications. First, minimizing this round's dilution at the cost of a lower next-round valuation is usually a bad trade (see `fundraising.md` — optimize the A, not the seed). Second, the *option pool refresh* is dilution too, and it's the one founders forget to model.\n\n**Model the option pool refresh at every round or you will be surprised twice.** Investors expect the pool to be topped back up to 10–15% at each round *[bench 2026-07]*, and expect it created **pre-money** — meaning existing shareholders eat all of it (see `startup-mechanics.md` for the mechanic). The refresh is not a one-time event; it recurs. A founder who models three rounds of price dilution but zero rounds of pool refresh is off by roughly a third on their final ownership.\n\n**Valuation is a multiple on ARR, and the multiple is set by growth, NRR, and gross margin — not by your pitch.** In practice, private software valuation is forward ARR times a multiple, and the multiple is a function of three things you control: growth rate, net revenue retention, and gross margin. A 100%-growth, 130%-NRR, 80%-margin company and a 40%-growth, 95%-NRR, 55%-margin company at the same ARR are not close in value, and no narrative closes that gap. If you want a higher valuation, the lever is one of those three numbers, not the deck.\n\n**ARR quality is priced separately from ARR growth — and AI-era revenue gets the quality discount.** The multiple attaches to revenue the buyer believes will still be there in three years. A diligence team will disaggregate your ARR into contracted-and-renewing, pilot-and-credit-funded, consumption above committed minimums, and services — and apply a different multiple to each, or exclude the weak lines entirely. A company reporting $10M ARR where a third is annualized pilots and a quarter is uncommitted usage is not valued like a $10M subscription business, it's valued somewhere near the durable core with a discount for having presented it the other way. Gross margin compresses the multiple through the same channel: inference COGS moves you from software comps toward something closer to tech-enabled-services comps. Do the disaggregation yourself before an investor does it for you — see `finance-ops.md` for the underlying metric definitions.\n\n**The Rule of 40 is the compression of the growth-versus-burn argument into one number.** *[bench 2026-07]* Growth rate plus profit margin (free cash flow margin, or burn as a negative) should sum to 40 or better. Its virtue is that it prices the tradeoff explicitly: 80% growth at −40% margin and 20% growth at +20% margin are the same score. Its limit is stage — at seed and early Series A it's meaningless noise, because a company going from $500K to $2M has an unbounded growth rate and a catastrophic margin. It becomes a real constraint somewhere north of $10–20M ARR, and it is how public comps get priced, which is how private multiples get set. Private SaaS medians have hovered near the high 20s in recent benchmark cycles, with top quartile around 48+ — but the benchmark moves with the rate environment, so check current data rather than trusting a number in a reference file. *(Stage: Series B+.)*\n\n**The waterfall, not the valuation, determines what you get paid.** Every dollar of preferred stock gets paid before a dollar of common. On a $200M exit with $80M raised in clean 1x non-participating preferred, the preferred converts to common and everyone shares pro rata — fine. On the same exit with a 2x participating preference, $160M comes off the top and *then* the preferred also shares in the remaining $40M. Same headline exit, radically different founder outcome. Build the waterfall model once, with real term sheets, and update it every round. Most founders build it for the first time during an acquisition, which is roughly two years too late. How to build it — which documents actually govern, how to read the stack series by series, and the prices at which each series converts — is in \"How to compute your waterfall\" below.\n\n**There is an exit range in which the founders get nothing and everyone else does fine.** The dead zone runs from zero up to the total preference stack: below it every dollar is spoken for and common gets exactly nothing. Above the stack common is paid, but at less than its ownership until the last series converts. A company that raised $150M and sells for $160M pays its investors and leaves the founders and employees with essentially zero, while the press release reads like a success. Know where your dead zone ends. It moves every time you raise, and raising more moves it *up*. This is the concrete, unglamorous reason \"raising too much\" is a real risk and not just investor scolding.\n\n**409A is not your valuation, and the gap is the point.** The 409A is an independent appraisal of *common* stock and typically comes in well below the preferred price — commonly somewhere in the range of a quarter to a half of it at early stage *[bench 2026-07]*, tightening as the company matures. That gap is legitimate: common lacks the liquidation preference, the protective provisions, and the information rights. The founder consequence: never tell a candidate their options are worth strike-times-headline-valuation. Tell them the strike price, the share count, the fully diluted total, and the preference stack ahead of them, and let them do their own math. Founders who inflate option value lose the trust of exactly the employees they most need at exit.\n\n**A down round is a balance-sheet event, not a moral one.** Founders resist down rounds as an admission of failure and take structure instead — participating preferred, ratchets, 2x preferences, pay-to-play — to preserve a headline number. This is almost always the worse trade. Structure is permanent, compounds through subsequent rounds, and is invisible to everyone except the people who will decide your outcome. A clean down round resets the price, refreshes the option pool at a strike employees will actually value, and is over. The market has repriced thousands of companies; it has not stopped funding any of the good ones.\n\n**Anti-dilution protection is why the *next* round's price matters this round.** Broad-based weighted-average anti-dilution (standard, acceptable) adjusts prior investors' conversion price if you raise lower later. Full-ratchet (rare, aggressive) reprices *all* their shares to the new low price — which can transfer enormous ownership in a single down round. This is the term that turns a bad quarter into a lost company. Check which one you signed. Most founders don't know.\n\n**Take secondary when it lowers your risk aversion, not when it looks like a win.** A founder with no personal runway makes bad decisions: they take the acquisition offer too early, they refuse the risky-but-correct pivot, they optimize for not-losing. Selling 5–10% of your position at a Series B or C to remove that pressure is usually good for the company and most decent investors will support it. The signal risk is real but overstated at modest sizes; it becomes a genuine red flag when a founder sells a large fraction, or sells while telling investors the company is about to inflect. Discuss it with your lead before you need it, not during the round. *(Stage: Series B+.)*\n\n**\"We'll fix it in the next round\" is a cost-of-capital decision you are making unconsciously.** Every deferred hire, deferred infrastructure investment, and deferred cleanup is implicitly financed by future equity — the most expensive capital available. Sometimes that's correct: speed is worth more than efficiency at the right moment. But make it a decision. The founders who get into trouble are the ones who never noticed they were financing operating decisions with the balance sheet.\n\n## How to compute your waterfall\n\nTwo axioms above say the waterfall determines the payout and that a dead zone sits under the preference stack. Neither says how to compute yours, and the corpus's standing complaint — that the model gets built during an acquisition — is really a complaint that nobody told founders what to open. What follows is the mechanics. The inputs are all knowable today and the arithmetic is grade-school; the reading of the documents is not, because liquidation preference is charter language, and what a given clause does (deemed-liquidation triggers, conversion mechanics, seniority) is a legal question with money attached. Build the model yourself, and have company counsel confirm the terms you read out of the charter before you act on any number it produces.\n\n**Your preference terms live in the certificate of incorporation, not the term sheet.** The term sheet was non-binding and got superseded at closing, the cap table spreadsheet carries share counts rather than terms, and the Carta screen shows ownership percentage — the number that stops being true the instant a preference is paid. The operative document is the amended and restated certificate of incorporation, restated at every priced round, so only the most recent one governs. Pull five things before you open a spreadsheet: that charter, the cap table with fully diluted shares by class, the option pool including unissued shares, every SAFE and note that has not yet converted, and the side letters. The side letters are the ones nobody remembers.\n\n**Read the stack series by series and write down four things for each.** The preference multiple (1x is standard; above it is structure you traded for a headline *[bench 2026-07]*); participating or non-participating; any cap on participation, written as a total multiple of invested capital; and seniority — whether the series are *pari passu* or stacked, which lives in the charter's liquidation article. Seniority is the one founders most often did not know they agreed to. On a $6M exit against a $2M seed and an $8M A: *pari passu*, the two split it in proportion to their preferences — $4.8M to the A, $1.2M to the seed. Stack the A on top and it takes all $6M; the seed gets nothing. One word, same charter.\n\n**Every non-participating series has a price above which it converts and stops taking preference.** Non-participating preferred takes the greater of its preference and its as-converted share — that \"greater of\" is the whole non-linearity. The first-pass conversion point is preference divided by as-converted ownership: an $8M Series A holding 20% converts above $40M, a $2M seed holding 20% above $10M. Then correct for the stack, because each series chooses against a pool every unconverted preference already shrank — with the A's $8M off the top, the seed's 2M shares are 25% of the 8M remaining, so it converts only where 25% of (exit − $8M) beats $2M: $16M, not $10M. Below every conversion point preferences come off the top; above the highest, the stack collapses to pro rata.\n\n**Participating preferred does not obey that formula, and uncapped it never converts.** Uncapped participation takes the preference *and* the as-converted share at every price, so converting is always the worse choice and there is no conversion point at any exit. A cap only moves where the arithmetic bites: the series converts where its as-converted share beats the cap — cap divided by as-converted ownership, not preference divided by it. Put a 3x cap on that same $8M Series A and it converts at $120M, not the $40M the non-participating formula gives. Run the wrong formula on a participating series and you compute the point three times too low.\n\n**Walk it at three prices, not one.** Take 6M common including the whole pool, plus a $2M seed and an $8M Series A at 2M shares each — 10M fully diluted, both 1x non-participating, *pari passu*. At $10M the preferences exactly consume the proceeds: $8M to the A, $2M to the seed, nothing to common. At $30M the A's 20% is worth $6M against an $8M preference, so it takes the $8M; the seed's alternative is 25% of the remaining $22M, or $5.5M, so it converts; common takes the other 75% — $16.5M, $2.75 a share, 55% of an exit in a company it owns 60% of. At $100M both convert and common takes its full 60%: $60M, $10.00 a share.\n\n**Find the price where common's first dollar arrives, and re-find it every round.** In the example that price is $10M — one dollar over the stack — and common does not reach its honest 60% until $40M, where the A is exactly indifferent and both routes through the model pay common the same $24M. Everything under $10M is the dead zone the axiom above describes. Raising moves both numbers unequally: a $20M Series B taking 20% adds its full $20M to the stack and pushes the last conversion point from $40M to $100M. That gap is the arithmetic behind \"raising too much is a real risk,\" and you compute it before signing the term sheet, not after.\n\n**The simple model gets the shape right and the number wrong.** An unconverted SAFE or note is not in your share count, and it converts at its cap into the round ahead of the exit, taking shares from common at a price nobody renegotiated. Cumulative dividends accrue onto their series' preference. A management carve-out comes off the top before the waterfall runs, negotiated rather than calculated. Counting the whole pool as outstanding common understates the per-share number: unallocated shares are cancelled at close, and issued options net their strike. And it all sits under a deal layer — escrow, earn-out, retention pool — that `exits-ma.md` covers and no cap table models. A real distribution runs off a payment spreadsheet prepared by company counsel; misallocating there is a claim from the shareholders you shorted.\n\n## How to run a down round\n\nThe down-round axioms above say what one is and which term decides what it costs. Neither says how to run one, and the sequence — who hears it first, what you concede, when the team is told — is where a survivable repricing becomes a distressed recap. What follows is the mechanics. A down round is a securities transaction and usually a conflicted one: it almost always requires amending the certificate of incorporation and stockholder consent, and when your existing investors lead it, the directors they appointed sit on both sides of the price — which Delaware, as of July 2026, reviews far more skeptically than an arm's-length deal. That is what makes board process a defense rather than a formality: disinterested approval or an independent committee, counsel in the room, contemporaneous minutes (`governance.md` on who holds which seat). Retain company counsel who has priced down rounds before; your lead's firm is not your counsel.\n\n**Your existing lead hears it from you, before the market and the team.** An investor who learns their mark is coming down from a co-investor, a banker, or another portfolio founder arrives as an adversary who also holds votes. Bring three things and nothing else: the honest number, the plan that number funds, and the specific ask — a bridge, their pro rata, a lead, or just non-obstruction. Naming the ask is the whole discipline, because opening with \"what do you think?\" hands them the pricing. Then prospective leads, then the team, never in parallel — parallel is how your team hears it from a recruiter. Find the exception first: a fund at the end of its life cannot follow at any price, which changes who you are negotiating with.\n\n**Walk in knowing which of your own terms the new price triggers.** Anti-dilution reprices earlier preferred's conversion automatically at close — no vote, no negotiation, and the anti-dilution axiom above owns which flavor you signed. Read three documents before the first meeting: the certificate of incorporation for the adjustment formula and the protective provisions, the investors' rights agreement for pro-rata rights, and the voting agreement for who elects which seat. The protective provisions are where the veto lives — a charter amendment and a new senior series generally need preferred approval as a class, sometimes series by series. Pay-to-play punishes only the investors who don't follow, which makes it your one real lever on a holdout. The founder who discovers a blocking right mid-negotiation has already lost the negotiation.\n\n**Compute what the headline costs before you trade structure for it.** The down-round axiom above owns the trade; here is the number. Take $30M at a $70M pre-money and the new investor owns 30% of a $100M post on clean 1x non-participating. Preserve a $200M headline instead — $170M pre, 15% — and pay for it with a 2x participating preference. At a $120M exit with earlier preferred converted, the clean deal pays the greater of $30M and 30% of $120M, so $36M, leaving $84M to everyone else. The structured deal takes $60M off the top, then 15% of the remaining $60M, so $69M — leaving $51M. The fifteen points the headline bought cost the rest of the cap table $33M.\n\n**Fight the option pool, not the headline — that is where points actually move.** Topping the unallocated pool from 5% to 15% pre-money takes ten points of a $100M post — $10M — out of your side before the round is priced, so the $70M pre-money you negotiated is a $60M pre-money. It arrives as housekeeping, sized by a hiring plan the new investor wrote, and never argued at the intensity price gets. Bring your own plan with named roles and grants; a pool sized to twelve months of actual hires is a negotiation, and one sized to a percentage is a concession. The refresh axiom above owns modeling the pool across rounds; this is how you negotiate it in this one, and `startup-mechanics.md` has why pre-money creation puts the whole cost on you.\n\n**Tell the team before close, and do their option math in the room.** Every employee will compute their grant against the new 409A, so get there first with three things: which grants are underwater — every strike above the new common fair market value — whether a repricing or refresh grant is coming and when, and what the money buys in months, because \"$30M against $2.5M of monthly burn is twelve months\" is a sentence people can hold. If none is coming, say so; a repricing is a board action with real tax and accounting consequences, sharpest for ISOs, and promising one you haven't cleared with counsel costs more than the underwater options do. The people who leave over a down round mostly leave over how they found out (`crisis-resilience.md`).\n\n**A down round is survivable and a slow one is not.** A priced round runs eight to twelve weeks from first conversation to wire, with the charter amendment and stockholder consents at the end of it *[bench 2026-07]* — so a process opened at six months of runway closes with three months in the bank, and one opened at four months closes with one. Every week of drift moves terms one direction only: leverage comes from a zero-cash date the other side already knows (`finance-ops.md`). That is what converts a repricing into a recap — same company, same price, worse structure. The limit: speed is not taking the first term sheet, and the fastest close available is usually the insider round, precisely the deal that needs the process above.\n\n## Common founder mistakes\n\n- Not knowing the preference stack. Ask any founder past Series A what their total liquidation preference is; a surprising number can't answer. It's the single number that determines whether an exit is life-changing or nothing.\n- Modeling dilution one round at a time instead of to exit, and omitting option pool refreshes entirely.\n- Quoting employees an option value based on the preferred price. It overstates the value, and it's the kind of overstatement people remember.\n- Taking venture debt to bridge a round they can't raise, rather than to extend into a milestone they can hit.\n- Accepting structure (participation, ratchets, multiple preferences) to protect a headline valuation. The headline is forgotten in a year; the structure is on the cap table until exit.\n- Billing monthly out of habit and never testing an annual-prepay discount — or testing it and reading the discount as an annual rate, which makes it look cheap next to equity.\n- Treating collected cash from annual contracts as though it were earned revenue, then discovering the deferred revenue liability during diligence.\n- Never building the waterfall model. It takes an afternoon with the term sheets and answers the question the whole exercise is about.\n- Building it from the cap table and the term sheet without ever opening the charter, where the multiple, the participation cap, and the seniority actually live.\n- Negotiating the pre-money hard and waving the option pool top-up through as housekeeping. Ten points of a $100M post leaves the founders' side before the price is set.\n- Opening a down-round process at four months of runway. Eight to twelve weeks plus consents means closing with one month of cash left, and the other side knows it.\n- Assuming the 409A is a company valuation and quoting it to anyone — press, candidates, or family.\n\n## What to read\n\n- Aswath Damodaran, [*The Dark Side of Valuation*](https://pages.stern.nyu.edu/adamodar/pdfiles/eqnotes/darksideyoung.pdf) — free NYU Stern lecture notes on valuing young, high-uncertainty companies. The rigorous layer under everything above.\n- Brad Feld & Jason Mendelson, *Venture Deals* — the standard reference on terms, preferences, and anti-dilution mechanics.\n- Bessemer Venture Partners, State of the Cloud, reissued each year on [Atlas](https://www.bvp.com/atlas) — annual benchmarks for multiples, efficiency, and the Rule of 40.\n- Christoph Janz, the SaaS Funding Napkin, published most years in [Point Nine Land](https://medium.com/point-nine-news) — what metrics each round actually requires. Find the newest one; an old napkin is a picture of a market that has moved.\n\nFile v1.9.0:references/cofounders-equity.md\n\n# Co-founders & Equity\n\n## Core axioms\n\n**The co-founder breakup is the #1 cause of startup death.** More than market, more than money, more than competition — people problems are the largest single category of venture failure, and the breakup is the one founders most reliably under-plan for. It is the most important and most ignored pattern in startup land.\n\n**You can optimize for rich or for king, but rarely both — and the founders who never choose get neither.** (Noam Wasserman, *The Founder's Dilemmas*.) Every consequential early decision prices the same trade of wealth against control: co-founder count, equity to hires, whose money you take, whether you'd accept a professional CEO. Founders who kept both the CEO seat and board control ended up with markedly less valuable stakes than those who gave up both, with giving up one landing in between — and the comparison is the founder's own slice, not the size of the company. The failure mode is refusing to pick: guard control at every turn, take venture money anyway, and you end up diluted *and* replaced. The escape hatches are narrow: enough leverage to set your own terms, or a business that never needs outside capital.\n\n**Vesting is the floor of the founder agreement, not the whole of it.** Four years with a one-year cliff, founders included, is the non-negotiable baseline — see `startup-mechanics.md` for the mechanics. But a clock only measures time, and a day-one split prices contributions nobody has made yet, so the gap between expected and actual contribution turns into resentment with no relief valve. The prescription is a dynamic agreement rather than a better opening guess (Noam Wasserman, *The Founder's Dilemmas*), and about half of founding teams build in no dynamic terms at all: no vesting, no buyout provisions, nothing. Name the machinery while everyone still likes each other — buyback terms, part-time treatment, revisit triggers — but keep it bounded: a split nobody can rely on buys flexibility by spending commitment.\n\n**Have the hard conversation early.** Roles, equity splits, decision rights, what happens if someone leaves — all on paper before you take the first dollar. Founders who delay this conversation because \"we're friends, we'll figure it out\" discover that figuring it out is exponentially harder once the company has any value. *(Stage: Pre-seed.)*\n\n**Schedule the transparent conversations on a recurring basis.** (Michael Seibel.) The topics that destroy co-founder relationships are the ones founders avoid: Is my co-founder working as hard as I am? Are we trying to accomplish the same goal? Whose job is engineering vs. product? Whose call is the next strategic decision? When these aren't surfaced, resentment builds. When resentment finally erupts, it does so as a fight rather than a structured discussion, and the relationship usually doesn't survive. The fix: a recurring scheduled time (weekly or bi-weekly) where you each name what's been bothering you. Boring meetings are the price of avoiding catastrophic conversations.\n\n**Assign your IP to the company at incorporation.** (Carolynn Levy / Jon Levy / Jason Kwon.) When founders buy their initial stock, the consideration usually includes assigning to the company any IP they've already created related to the business — and IP the company never formally took ownership of is what diligence finds. See `startup-mechanics.md` for the full treatment.\n\n**Equal equity splits are usually wrong but often right anyway — what should worry you is how fast you got there.** The math says it should be unequal — different contributions, different opportunity costs, different ongoing roles — but the relationship usually requires equal anyway, because resentment is more expensive than a few percentage points. Yet 42% of founding teams settle the split in a day or less, and teams that negotiate quickly and land on even shares grow headcount more slowly and raise venture money less often (Thomas Hellmann and Noam Wasserman, \"The First Deal\"). The 50/50 isn't doing the damage; the speed is — a symptom of a team avoiding exactly the conversations it has to be able to survive. Decide deliberately, and whatever you land on, write down why.\n\n**The CEO is the CEO.** A 50/50 split with no clear decision-maker means decisions don't get made. Every co-founder pair needs one person designated as the tiebreaker — usually called CEO — and the other co-founders need to genuinely agree that this person has final say on contested decisions. Otherwise the company freezes at every disagreement.\n\n**Co-founder roles should be non-overlapping.** Two co-founders both \"doing product\" is a recipe for daily friction. Even when skill sets overlap, draw a sharp line: one owns product, the other owns engineering. One owns sales, the other owns operations. Ambiguous ownership creates conflict in proportion to the importance of the area.\n\n**Everyone can't be C-something.** (Noam Wasserman, *The Founder's Dilemmas*.) The CEO seat is a decision-rights question, settled above; CTO, CPO, and COO are titles, handed out by domain rather than by proven ability to run the function at scale — on day one there is nobody to be chief of. They are close to impossible to downgrade later: the day you need a real VP of Engineering, you are negotiating a title away from your co-founder instead of making a hire — a negotiation about status, not the role. Give plain functional titles early — head of engineering, head of product — and let executive titles be earned by people who have run the function at the size you are about to be (see `management-execution.md`).\n\n**The \"ideas guy\" with no skills is not a co-founder.** Ideas are cheap; execution is everything. A co-founder who can't write code, sell, design, or operate is not a co-founder — they're an advisor at most. Resist the social pressure to make a friend a co-founder because they were there for the brainstorm.\n\n**Get double-trigger acceleration into your founder agreements — having no acceleration is the mistake.** Double-trigger (you vest fully on acquisition *plus* involuntary termination) is the market standard and what to secure: it protects you from being acquired and then fired out of your unvested shares. Single-trigger (full vesting on acquisition alone) sounds better but acquirers and investors resist it — it removes the retention the buyer is paying for, and pushing for it can cost you elsewhere in the deal. Founders with no acceleration at all end up working for the acquirer for years they didn't sign up for.\n\n**Co-founder therapy is real, useful, and cheap insurance.** A monthly session with a coach or therapist explicitly focused on the co-founder relationship — not personal therapy — prevents most of the conflicts that destroy companies. Most founders won't do this until things are bad. The ones who do it pre-emptively almost always say it was the highest-ROI thing they did.\n\n**Equity splits look small until they don't.** A 60/40 vs 50/50 split feels like a small choice at incorporation. At a $500M exit, it's a $50M difference. Take the decision seriously the first time so you don't have to renegotiate it under duress later.\n\n**Give up n% of the company only if the deal makes the whole company worth more than 1/(1−n) times what it was.** (Paul Graham, \"The Equity Equation.\") One test covers every equity trade — investor money, a hire, a partnership: the (100−n)% you keep has to beat the 100% you held, measured against the company's *average* outcome, not its headline valuation. Inverted for hires — where *i* is how much you believe someone multiplies that average outcome — the break-even grant is (i−1)/i: believe they lift it by a fifth and you break even at 16.7%, a ceiling and not an offer, since salary and overhead come out of the same stock. The limit is Graham's own: the multiple is a guess, and the equation's value is that it names what you are guessing at. *\"You should always feel richer after trading equity.\"*\n\n**The founder agreement is not the cap table.** Operating agreements, founder agreements, and vesting documents are separate from the cap table — and they're where the real protection lives. IP assignment, non-competes, mutual buyout rights, dispute resolution: all of this belongs in the founder agreement, signed before you raise.\n\n**You don't need a co-founder; you need the right co-founder.** Solo founders raise less money — but a wrong co-founder is far worse than no co-founder. Don't grab a co-founder out of YC pressure or to seem more legitimate to investors. Run alone until the right person appears. *(Stage: Pre-seed.)*\n\n**Co-founding with friends or family is playing with fire; the stable base is people you have actually worked with.** (Noam Wasserman, *The Founder's Dilemmas*.) Socially-bonded teams are more than half of all founding teams and the least stable kind: higher founder turnover than former-coworker teams, with acquaintances in between. Friends treat the knowing as already done, skip the assessment two acquaintances would run as a matter of course, and are the least likely of any team type to put conflict-loaded topics on the table. The downside is worst here: when it breaks, it takes the relationship along with the company. The variable is not how well you know someone but whether you have worked with them — so run the diligence you would run on a stranger, write the agreements, and force the conversations the friendship is protecting you from.\n\n**The co-founder breakup almost always has signs you ignored.** Resentment that simmers for a year doesn't appear suddenly. The \"I want out\" conversation usually follows 6–18 months of avoided difficult conversations. Catching this early — through honest 1:1s, through co-founder therapy, through a board member or investor who'll mediate — saves the company.\n\n**Interviewing a CTO co-founder is like getting married off Bumble.** (Phil Green — an unrecorded talk; no published source located, so what follows paraphrases him rather than quoting him.) The CTO/co-founder relationship is 5–10 years of 8-hour days through thick and thin — you cannot select for that in two coffee meetings. Start them as a technical lead or contractor; let the co-founder conversation emerge from working together for three to six months. Many of the co-founder breakups that destroyed companies could have been avoided by running this kind of audition before committing equity. *(Stage: Pre-seed.)*\n\n## Common founder mistakes\n\n- Settling the split in one conversation and shaking on it. The 50/50 is rarely the damage — the speed is, per the equal-splits axiom above — and a handshake leaves you with no vesting and no buyout terms on the day one of you wants out.\n- Letting a co-founder coast post-product-market-fit. The early-stage hero who built the MVP often doesn't have the skills to operate at scale. Founders who let an underperforming co-founder ride along destroy team morale.\n- Bringing in a co-founder months into the journey at equal equity. The person who showed up six months later did not take the same risk and should not get the same equity. Use a vesting schedule that starts from their join date, not the company's formation date.\n- Promising a co-founder-sized stake in conversation and papering it later. The promise is not a grant: what eventually lands is set by a board approval and a 409A on the company's schedule, not by the number the two of you remember agreeing to. See `startup-mechanics.md` for the documentation mechanics.\n- Avoiding the conversation about what happens if one co-founder wants to leave. Every founder agreement needs a buyout provision — what's the price, who triggers it, what's the timeline.\n\nFile v1.9.0:references/crisis-resilience.md\n\n# Crisis & Resilience\n\n## Core axioms\n\n**Not dying is most of the strategy.** Graham said it to a YC batch — \"If you can just avoid dying, you get rich\" (Paul Graham, \"How Not to Die,\" 2007) — and the population is the point: in an already-filtered cohort, enough companies succeed that survival and success nearly coincide. Every startup is 30 days from death at some point; even the most successful companies have stories of being weeks from missing payroll, and the differentiator is never avoiding the moment, it's the founders who keep showing up through it. That's why the near-death moment is survivable and the slow fade — the founders quietly giving up — generally isn't. See `yc-canon.md` for the full treatment of why the binding constraint is usually founder commitment, not cash. The discipline is interpretive: the low points run lower than anyone warns you, and reading a trough as a verdict is the expensive mistake. The limit is the filter itself — for a company with no path, the same posture turns a clean wind-down into a slow bleed.\n\n**Bad news doesn't age well.** Whatever you're avoiding telling your team, investors, or board — tell them today. The delay never makes the news easier to deliver, but it does usually make it worse. Founders who develop the discipline of \"name the problem within 48 hours of recognizing it\" build trust in proportion to the difficulty of the news they share. That's the outbound discipline; whether bad news ever reaches you is set by how you react when it does — see `culture.md` on how every reaction to bad news trains the company what to tell you.\n\n**Layoffs done right are deep and once.** Two rounds of layoffs in six months destroys trust. Cut deeper than feels comfortable, then commit: cut once, cut deep — and the rule is the same for burn and scope as for headcount, at any size. The founder instinct to \"give it another quarter\" before a hard cut almost always costs more in the end, on brutally simple math: if you're losing money, every quarter of delay is real cash that won't be there when you need it. The team that survives a single decisive layoff can refocus; the team that suffers serial layoffs becomes a job-hunting team in slow motion. A layoff is a different event from a performance exit — it's about the plan, not the person, and saying so plainly is what preserves the departing person's next job — but the execution mechanics are the same: same-day access, prepared severance, a single announcement. See the \"How to fire someone\" section in `hiring.md`, and get counsel on mass-layoff notification rules before you set a date. *(Stage: Series A+ for the layoff mechanics; cut-once-cut-deep on burn and scope holds at any headcount.)*\n\n**The CEO sets the emotional thermostat.** Your team will mirror your anxiety. Calm in a crisis is a learnable skill and an operational asset. This doesn't mean false optimism — it means clear-eyed acknowledgment of the situation paired with a credible plan and visible composure.\n\n**Peacetime CEO vs. Wartime CEO.** (Ben Horowitz.) A founder's management style must adapt to the strategic context — and the failure mode is using the wrong one for the moment. *(Stage: Series A+.)*\n\nThe **Peacetime CEO** operates when the company has a clear competitive advantage and the work is to scale the opportunity. Long-term planning, robust systems, minimal internal conflict, broad-based creativity, deference to protocol. Decisions made by consensus; disagreements aired and reconciled. Investing in culture, training, and managers because the company is around for the long run.\n\nThe **Wartime CEO** operates when the company faces an existential threat — imminent cash-out, a competitor moving on your core market, a market shift that could obsolete you. No time for consensus. Extreme operational focus, intolerance of deviation from the plan, central control over minutiae that affect survival. The wartime CEO heightens contradictions, bypasses protocols to win, and lets the crisis define the culture for the duration.\n\nHorowitz's point is that **the same person often needs to do both**, and the transition is hard. Applying peacetime methods in wartime — collecting input, deliberating, respecting protocol — can sink the company. Maintaining wartime methods in peacetime — micromanaging, heightening conflict, suspending normal decision rights — burns out the team and stifles the innovation that powered the company's success in the first place. Founders who can't switch modes either get fired during wartime (their boards lose patience) or lose their best people during the peacetime that follows a crisis (their team is exhausted from being managed in war when there's no war).\n\nThe signal you've entered wartime: the question is no longer \"how do we grow faster\" but \"do we survive the next quarter.\" The signal you've returned to peacetime: the existential threat is genuinely behind you, not merely deferred. Misreading either signal is expensive.\n\n**Crisis converts a correct diagnosis into a mandate.** (Brian Chesky, Lenny's Podcast, 2023.) Most founders can see what needs resetting long before they are allowed to reset it: the binding constraint is the organization's consent, not the founder's analysis. An existential threat removes it — during COVID the same people at Airbnb who had resented Chesky's involvement started demanding it. What Horowitz's wartime mode above describes as a management style, the organization on the receiving end experiences as permission. The limit is that same line: the threat licenses a reset, not a standing suspension of decision rights, and holding the wartime posture past the wartime burns out the team that just carried you through it. See `management-execution.md` for the argument about how involved a founder should be in normal times. *(Stage: Series B+.)*\n\n**In a crisis, communicate 3x more than feels necessary.** The team in the absence of information will invent their own narrative, and the invented narrative is almost always worse than reality. Weekly all-hands during a crisis, weekly written updates, open Q&A — over-communication is the only safe default.\n\n**The bridge round, the layoff, and the pivot are usually all the same conversation, six weeks apart.** When the business is in trouble, founders often try to fix it with a small intervention, then a bigger one, then a structural one — each separated by weeks during which the trouble deepens. The pattern recognition is to telescope the conversations: assume the problem is bigger than it first appears and act accordingly.\n\n**Pay attention to the people who stay.** During and after a crisis, focus disproportionate energy on the people who remain committed. They're watching how you behave. The team that stays through a hard year is the team that wins the next year — but only if they feel the founder treats them as partners, not as survivors of the founder's mistakes.\n\n**Acknowledge mistakes, then move.** Founders who refuse to acknowledge bad calls lose the team. Founders who marinate in their bad calls also lose the team. The pattern: name the mistake, explain what you learned, describe what you're doing differently, then act. Same conversation, 10 minutes, then move forward.\n\n**The decision to shut down is harder than the decision to fight on, even when shutdown is right.** Most founders fight too long. The signal: when continuing requires increasingly unrealistic assumptions to justify, it's time to wind down. The grace and discipline to do this — returning capital to investors, helping the team find roles, leaving the door open for future ventures — defines founders who go on to do bigger things later. That's the whether; the how is a separate discipline, and the sequence is in \"How to wind down\" below.\n\n**Customers are the last to know they should leave you.** During a company crisis, your customers usually have no idea. Don't lie about it, but also don't volunteer your panic. The instinct to \"be transparent\" with customers about runway problems often accelerates churn unnecessarily. Tell them what they need to know to make their decisions — no more.\n\n**Founders carry the weight; founders need a release valve.** A weekly call with another founder, a coach, a therapist, a long run with no phone. The founder who has no release valve breaks. The breakdown comes in different forms — bad decisions, lashing out at the team, paralysis, health problems — but it comes.\n\n**You are not your company.** The startup that fails is not a referendum on the founder's worth. This is hard to feel when you're in it, but it's true. The data: most successful founders had failed ventures before the successful one. Failing is a phase, not an identity.\n\n**The post-mortem matters most when you don't want to do one.** Companies that fail often skip the honest post-mortem because the founders are exhausted and demoralized. The honest post-mortem — written, shared, true — is the input to the next venture. Skip it and you'll repeat the mistakes.\n\n**Resilience is not endurance; it's recovery.** The founder who pushes through a crisis without ever recovering breaks. The founder who knows when to step back, when to take a week off, when to ask for help — that founder is the one who lasts. Endurance is the wrong metaphor; the right one is interval training.\n\n**A failed project's most valuable output is often reusable tech, not the post-mortem.** (Alphabet X.) When Loon (balloon internet) died, the team's discipline was \"never throw anything away\" — the optical-link tech built to connect balloons became Taara, now a standalone business. When a project fails on unit economics but the underlying technology works, the assets are the spinout option, not a write-off. Run the kill decision and the asset-harvest decision in parallel; don't let exhaustion collapse them into a single \"shut it down\" conversation. *(Stage: Series B+.)*\n\n## How to wind down\n\nThe axiom above says when to shut down, and `finance-ops.md` names the cash band where the decision stops being yours. Neither says how, and the gap between deciding and executing is where a company failure turns into a personal one — `startup-mechanics.md` inventories the standing exposures that make that conversion possible, and this is the sequence in which you meet them. What follows is the mechanics. Wind-down law is jurisdiction-specific and unforgiving in the same way employment law is — final-pay timing, accrued-PTO payout, creditor-notice requirements, and dissolution procedure are set by state law and vary, and mass-layoff notification (the federal WARN Act and its state equivalents) carries employer-size thresholds and notice periods that a shutdown can trigger just as a layoff does. Retain insolvency counsel while you can still pay them and have them run the sequence; a week of a specialist's time is the cheapest line item in the exercise and the first one founders cut.\n\n> **Law here carries a vintage.** This section reflects U.S. federal law and general state-law practice as of July 2026, with the fiduciary-duty framing drawn from Delaware. The mechanics are durable — remit the withholding, stop the meter, pick the right instrument — but the thresholds, deadlines, and procedures are not, and several of them differ state to state. Verify with counsel before acting on any of it.\n\n**Once you're insolvent, paying out of order always costs someone — the question is whether it costs you.** Secured creditors have first claim on their collateral. Among the rest, employee wage claims and withheld payroll taxes come ahead of general unsecured — though the wage priority is capped per person and covers only wages earned shortly before the filing. Equity is last and usually gets nothing. What founders do instead is pay on sentiment, and the two cases sentiment produces are not the same exposure. Paying an arm's-length creditor early — the freelance designer who invoiced last month, the small agency that will actually feel the loss — is generally within a Delaware director's discretion, and if a bankruptcy follows, the trustee's preference claim runs against the transferee: it is the designer who gets sued for the money back, months later, having long since spent it. That wastes the estate and burns the relationship, but it is not by itself a claim against you. The protection there is procedural rather than automatic: what earns it is deciding disinterestedly, in good faith and on an informed basis, with the board and counsel actually in the room — which is the part a cornered founder skips (`startup-mechanics.md`). Repaying your own founder loan forfeits it outright, because an insider transfer runs on a one-year look-back rather than ninety days and puts you on both sides of the payment. The limit: the priority scheme in a formal proceeding is statutory and fact-specific — counsel runs the waterfall, you don't improvise it. (Why the insider payment becomes a claim against you — the duty mechanics near insolvency — is in `startup-mechanics.md`.)\n\n**Withheld payroll taxes are not your money, and they follow you home.** What you withheld from employees' paychecks was held in trust for the government, not borrowed from it. Failing to remit is a personal liability that survives the company and generally survives your own bankruptcy. Remit before anything else, including your own salary, and read a payroll you cannot fully fund as the signal the company is done rather than as a financing decision. The full exposure is in `startup-mechanics.md`.\n\nThere are three ways to end a company, and only one of them leaves you holding the wheel.\n\n**Dissolution is the clean case and only the clean case.** The company can pay what it owes, so the board and stockholders approve, you file in your state of incorporation, give creditors the statutory notice, and distribute what's left. You keep the wheel the whole way.\n\n**An assignment for the benefit of creditors is the venture default when the company is insolvent but the failure is uncontested.** The board, usually with stockholder approval, hands the assets to a professional assignee, who liquidates and distributes them in priority order — months, not weeks, and at a fraction of bankruptcy's cost. The asset sale can close fast; the claims bar and the distribution cannot. It's the usual vehicle for selling the IP or landing the team, but it is a creature of state law: availability, court involvement, and the claims timeline vary substantially, and Delaware rewrote its version in 2026. Once you have named the assignee, you are a spectator.\n\n**Chapter 7 is for the contested case.** What puts you there: litigation only the automatic stay will halt, creditors who won't stand down, nobody willing to take the assignment. It hands everything to a trustee whose job includes examining what you did on the way down. What founders wrongly assume: none of the three erases a personal guarantee (`bootstrapping.md`) or the trust-fund liability above, and a corporation in Chapter 7 is liquidated, not discharged.\n\n**Stop accruing liabilities the day you decide, not the day you announce.** Between the decision and the shutdown email the cloud bill keeps metering, the lease keeps running, the contractor SOWs keep billing, and the annual contracts march toward renewal — every dollar of it an unsecured claim against the pot that was supposed to cover final payroll. Auto-renewal is the specific killer, because cancellation windows are written to be missed and a renewal that trips two weeks after you decided is a full additional year you took on knowingly. Work the list the same week you decide: every vendor, subscription, and commitment sorted by notice window and renewal date, cancelled in that order, before anyone outside the room knows. The line you cannot cross is ordering services you already know you cannot pay for.\n\n**Shut down while you can still afford to shut down well.** An orderly wind-down is a purchase, and the price is final payroll, accrued PTO where state law requires it paid out, any notice period law or contract imposes, counsel, the accountant who files the final returns, filing fees, and a few months of a registered agent and an open bank account to catch the last invoices. Founders spend straight through that floor chasing a last term sheet. Below it, the choice is no longer between a good shutdown and a cheap one — it's between a shutdown and a mess that settles onto your personal balance sheet, because the obligations don't disappear when the cash does. Model the floor in the 13-week and hold it as a reserve rather than counting it as six more weeks of runway (`finance-ops.md` names the same moment from the cash side). The tell you've already crossed it: every version of the plan requires the round to close.\n\n**You owe employees, customers, and investors in that order, and founders invert it.** Employees get the truth on the earliest date you can say it without making it a rumor, the severance the floor bought them and whatever more the creditors leave, references offered in advance rather than on request, and an introduction list you work personally. The market reads a shutdown very differently from a firing, and you're the only one positioned to say so. Customers get a real migration window and a usable data export. The axiom above about not volunteering your panic expires the moment the decision is final, and a 72-hour warning converts your failure into their outage. Investors get the letter — what happened, what the money bought, what you learned — and whatever survives the creditors, usually nothing. The inversion is the failure mode: the letter gets drafted first because it's the audience the founder most fears, and the team is told last because it's the conversation the founder is most ashamed of.\n\n## Common founder mistakes\n\n- Hiding the crisis from the team until it's so obvious that the cover-up itself becomes a crisis. Trust, once broken, is very hard to rebuild.\n- Treating the crisis as the founder's secret to bear. Co-founders, key executives, and the board all need to know. Carrying it alone produces worse decisions.\n- Layoffs designed to minimize discomfort for the layoff-er. Decisions made to make the founder feel better — spreading layoffs across multiple rounds, not telling the team you're considering layoffs until the day, choosing the easiest people to cut rather than the right ones — all create worse outcomes for everyone.\n- Refusing outside help. Crisis is when bringing in an experienced operator, interim executive, or turnaround advisor pays the most. Founders who view this as defeat lose more than founders who view it as professional execution.\n- Confusing perseverance with stubbornness. Perseverance is \"this is hard but the path is sound.\" Stubbornness is \"this is hard and I won't admit the path is wrong.\" The line is subtle but consequential.\n- Funding one more payroll cycle by skipping the withholding remittance and calling it a bridge. That liability is personal, survives the company, and generally survives your own bankruptcy.\n- Paying the small, sympathetic vendors first on the way down. It feels like decency, and a trustee can claw it back out of the vendor's hands months later — the person you were trying to protect is the one who gets sued for it. Deciding it on your own, without the board or counsel, is what turns it into your problem rather than the vendor's. The out-of-order payment that lands on you personally no matter how it was decided is the one you make to yourself: repaying the founder loan ahead of the trade creditors, on the longer insider look-back.\n- Spending through the wind-down reserve while waiting on a term sheet. The cheapest possible shutdown still costs real money, and the founder who can no longer buy one absorbs personally what the company can't pay.\n\nFile v1.9.0:references/culture.md\n\n# Culture & Early Team DNA\n\nCulture is one of the most ignored topics at the early stage — founders treat it as something to \"figure out later\" — and one of the most consequential. The behavior patterns set by the first 10–20 employees compound forever. This file draws heavily on Tim B\n\nArchive v1.0.1: 20 files, 87838 bytes\n\nFiles: CONTRIBUTING.md (6782b), LICENSE (1084b), README.md (7285b), references/cofounders-equity.md (7097b), references/crisis-resilience.md (8635b), references/culture.md (8135b), references/customers-market.md (10651b), references/finance-ops.md (8614b), references/fundraising.md (9291b), references/governance.md (6752b), references/hiring.md (7979b), references/meta.md (9687b), references/product.md (13361b), references/sales-gtm.md (10305b), references/startup-mechanics.md (12184b), references/time-energy.md (7377b), references/yc-canon.md (33271b), skill-card.md (2979b), SKILL.md (17050b), _meta.json (133b)\n\nArchive v1.0.0: 20 files, 79320 bytes\n\nFiles: CONTRIBUTING.md (6782b), LICENSE (1084b), README.md (6398b), references/cofounders-equity.md (6599b), references/crisis-resilience.md (8078b), references/culture.md (6183b), references/customers-market.md (8346b), references/finance-ops.md (8080b), references/fundraising.md (6644b), references/governance.md (6752b), references/hiring.md (6980b), references/meta.md (7686b), references/product.md (11642b), references/sales-gtm.md (7831b), references/startup-mechanics.md (12184b), references/time-energy.md (6800b), references/yc-canon.md (33271b), skill-card.md (3306b), SKILL.md (14969b), _meta.json (133b)","readmeExcerpt":"Skill: founder-wisdom Owner: chris-graffagnino Summary: Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, and crisis. Use whenever someone is wrestling with a startup, scale-up, or early-stage company decision — hiring or firing, picking a co-founder, whether to raise at all, bootstrapping and staying indepen","codeSnippets":[],"executableExamples":[{"language":"bash","snippet":"python3 scripts/build_bundle.py"},{"language":"bash","snippet":"python3 scripts/build_bundle.py --check"},{"language":"text","snippet":"founder-wisdom/\n├── SKILL.md                          # The main routing logic for Claude\n├── evals/                            # Behavioral scenarios that pin down SKILL.md\n│   ├── scenarios.yaml\n│   ├── check_scenarios.py\n│   └── README.md\n├── scripts/\n│   └── build_bundle.py               # Generates dist/ from SKILL.md + references/\n├── dist/                             # Generated artifacts for non-Claude LLMs (do not edit)\n│   ├── founder-wisdom-full.md\n│   └── system-prompt.md\n└── references/\n    ├── hiring.md\n    ├── fundraising.md\n    ├── product.md\n    ├── sales-gtm.md\n    ├── finance-ops.md\n    ├── capital-valuation.md\n    ├── bootstrapping.md\n    ├── exits-ma.md\n    ├── cofounders-equity.md\n    ├── governance.md\n    ├── time-energy.md\n    ├── customers-market.md\n    ├── crisis-resilience.md\n    ├── culture.md\n    ├── startup-mechanics.md\n    ├── strategy-moats.md\n    ├── management-execution.md\n    ├── socratic-technique.md\n    ├── yc-canon.md\n    ├── yc-canon-product.md\n    └── meta.md"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: founder-wisdom\ndescription: Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, and crisis. Use whenever someone is wrestling with a startup, scale-up, or early-stage company decision — hiring or firing, picking a co-founder, whether to raise at all, bootstrapping and staying independent, non-dilutive capital, selling the company, moats and positioning, how AI changes the way a startup team works, sells, and hires, who stays accountable when an agent does the work, or any \"I'm a founder and I'm not sure what to do\" moment. Trigger without an explicit ask for \"advice\" — \"should I…\", \"is it normal that…\", \"we're thinking about…\", and \"how do experienced founders handle…\" are in scope, as is helping someone else with founder-stage decisions (advisor, coach, mentor, investor). Also trigger on \"who said X\" questions about the startup canon.\nmetadata:\n  version: \"1.9.0\"\n---\n\n# Founder Wisdom\n\nA reference skill for surfacing the axioms experienced startup founders know in their bones — the pattern-matched wisdom that first-timers usually learn by running into walls.\n\n## What this skill is for\n\nFounders and operators face decisions where the right answer isn't obvious from first principles — it's obvious from pattern. \"Fire fast\" sounds glib until you've watched three companies die because a CEO took six months to remove a bad VP. \"Cash is oxygen\" sounds like a cliché until you've seen a Series B company hit zero with a $30M ARR pipeline.\n\nThis skill captures those patterns as axioms, organized by domain, with the context that makes each one useful. It is opinionated and pattern-matched, not neutral or comprehensive. Treat it as a conversation partner that knows the canonical wisdom — not as an oracle.\n\n## Two modes of operation\n\nThe skill operates in **direct mode** by default and **Socratic mode** when the conversational signal calls for it. Most uses will be direct.\n\n### Direct mode (default)\n\nWhen someone asks a clear question — \"what should I know about firing my first executive?\" or \"how do experienced founders handle a down round?\" — surface the relevant axioms directly. Lead with the axiom, then the reasoning, then the qualifier. Example:\n\n> **Hire slow, fire fast.** Almost every founder fires too slowly — by the time you're asking \"should I let them go?\", the answer is yes and was yes two months ago. The exception: never fire in anger, and never on a Friday.\n\nPull 3–7 axioms maximum per response. Seven is a hard ceiling, not a target — count the bolded lead-ins before you send, and if there are eight, cut to the ones carrying the answer rather than trimming each one shorter. More than that and you're dumping a list instead of giving advice. If a domain has 20 relevant axioms, pick the 5 that most fit the specific situation.\n\n**Bold marks an axiom and nothing else.** Not a step in a procedure, not a sub-point under an axiom you al"},{"path":"evals/README.md","content":"# Evals\n\n`SKILL.md` is prose that steers a model. Prose drifts. This directory holds a small,\nhand-checkable set of scenarios that pin down what the skill is supposed to do:\nwhether it triggers, which reference files it reads, which mode it picks, and\nwhether the output obeys the discipline in `SKILL.md`.\n\nThe directory has two halves. `check_scenarios.py` validates the data file's shape\nand calls no model — stdlib only, no key, runs anywhere Python does.\n`run_scenarios.py` executes the scenarios against a model and needs the `anthropic`\nSDK plus an `ANTHROPIC_API_KEY`. `scenarios.yaml` is the source of truth for both,\nand it is still a file you can read as documentation and run by hand.\n\n## Files\n\n- `scenarios.yaml` — the eval set. 38 scenarios plus five global invariants.\n- `check_scenarios.py` — stdlib-only validator and pretty-printer. It checks the\n  data file's shape, that every referenced path exists, and that `SKILL.md`'s\n  frontmatter description is still inside its 950-character budget (the runtime\n  ceiling is 1024; the gap is deliberate headroom, and the way to get it back is\n  to move enumeration into the routing list, not to raise the budget). It does\n  **not** call a model.\n  It bundles a minimal YAML reader rather than taking a PyYAML dependency; its output\n  on `scenarios.yaml` has been diffed against PyYAML and is identical.\n- `run_scenarios.py` — the executable half. It imports `load` and `validate` from\n  `check_scenarios.py` rather than reparsing the YAML, then drives each `prompt`\n  through a harness that hands the model one skill and two tools, so triggering and\n  file routing are read off the tool calls instead of inferred from the prose.\n  Deterministic checks always run; `--judge` adds a second model call that grades the\n  mode and the prose assertions. Needs the `anthropic` SDK and `ANTHROPIC_API_KEY`\n  unless `--dry-run`.\n- `results/` — committed output from judged runs. See \"Committed results\" below.\n\n## Schema\n\nTop level:\n\n| Key | Meaning |\n|---|---|\n| `version` | Schema version. Currently `1`. |\n| `global_invariants` | Plain-English checks that apply to every scenario where `should_trigger` is true. |\n| `scenarios` | The list below. |\n\nEach scenario:\n\n| Key | Type | Meaning |\n|---|---|---|\n| `id` | string | Stable kebab-case identifier. Referenced in PR discussion. |\n| `prompt` | string | Exactly what the user says. Verbatim — do not paraphrase when running. |\n| `should_trigger` | bool | Whether the founder-wisdom skill should activate at all. |\n| `expected_mode` | `direct` \\| `socratic` \\| `none` | `none` iff `should_trigger` is false. |\n| `expected_files.must_include` | list | Reference files that must appear among those read. Never more than three. |\n| `expected_files.must_not_include` | list | Reference files whose presence is a failure. |\n| `assertions` | list | Plain-English checks on the response. |\n| `rationale` | string | One line naming the `SKILL.md` rule the scenario protects. |\n\n`expected_files` is deli"},{"path":"README.md","content":"# Founder Wisdom\n\nA [Claude Skill](https://www.anthropic.com/news/skills) that surfaces hard-won axioms from experienced startup founders and operators — the pattern-matched wisdom that first-timers usually learn by running into walls.\n\n## What it does\n\nFounders face decisions where the right answer isn't obvious from first principles — it's obvious from pattern. This skill gives Claude access to a curated corpus of those patterns, organized by domain, with the context that makes each one useful.\n\nWhen you ask Claude a founder-stage question — hiring, fundraising, product, sales, finance, co-founders, governance, crisis, or any of the surrounding territory — Claude consults this corpus and responds with the relevant axioms rather than working from generic priors.\n\nThe skill operates in two modes:\n\n- **Direct mode** (default): Surfaces relevant axioms with reasoning and qualifiers. Useful when you want a quick read on a decision.\n- **Socratic mode**: Asks the question the axiom answers, rather than handing you the answer. Useful for working through ambivalence, for coaching/mentoring conversations, and for processing live decisions.\n\n## Domains covered\n\n- Hiring, firing, comp, equity grants, the keeper test, and the AI-era talent shift toward systems thinkers\n- Fundraising, terms, runway, investor relations\n- Product, product-market fit, pivots, and where startup ideas come from — Paul Graham on noticing rather than inventing, and on schlep blindness\n- Sales, pricing, GTM, the founder-as-seller, what a visionary customer is actually buying, and selling in the AI era — building the motion like a product, automating in order of workflow legibility\n- Finance, cash, burn, CFO timing, ramen profitability, AI-pilot revenue and ERR, the operating metrics that matter (cohorts, NRR, Rule of 40, CAC payback, gross margin), and how to build the two models the axioms keep pointing at — the 13-week cash flow forecast and the Default Alive calculation\n- Capital and valuation — cost of capital, dilution math, venture debt, ARR multiples, the liquidation waterfall, 409A, down rounds, secondaries, plus the procedures: computing your own waterfall off the charter, and running a down round\n- Bootstrapping and the non-venture path — whether to raise at all, self-financeable growth, overtrading, annual prepay, revenue-based financing, SBIR grants, profit share and phantom equity, the plateau\n- Exits and M&A — running a sale process, bankers, LOIs and exclusivity, diligence, earn-outs, escrow, retention packages, acqui-hires, headline price vs. actual payout\n- Co-founder dynamics, splits, vesting, titles, and Noam Wasserman's rich-versus-king trade\n- Boards and governance — board design, running the board like a team you lead, and why success is what gets founder-CEOs replaced\n- Founder time, energy, sustainability, and where founder attention goes — the maker's schedule, the one top idea in your mind\n- Customers and market dynamics, and Geoffrey Moore's chasm — why ea"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn7bwxjecdr5xfbq5tf88jsya583f0sv\",\n  \"slug\": \"founder-wisdom\",\n  \"version\": \"1.9.0\",\n  \"publishedAt\": 1786553282130\n}"},{"path":"references/bootstrapping.md","content":"# Bootstrapping & the Non-Venture Path\n\nMost of this corpus is calibrated to companies that raise institutional money, because that is where most of the observed pattern comes from. This file is the counterpart to that material rather than a rebuttal of it: the venture axioms hold for venture companies, and what breaks here is the subset that quietly assumes a round standing behind them. What changes is the binding constraint — growth is rate-limited by margin instead of by a fundraising calendar, nobody outside the company grades the number, and the characteristic death is a decade-long plateau rather than a zero-cash date. `capital-valuation.md` prices the instruments; this file covers what the absence of a round does to everything else.\n\nThis file is not financial, tax, or legal advice. The math is here so founders can ask their CFO and counsel better questions, not so they can skip them.\n\n> **Benchmarks are calibration, not truth.** Figures marked *[bench 2026-07]* — the revenue ceiling that separates a good company from a viable venture asset, annual-prepay discounts, revenue-based-financing caps and remittance rates, SBIR award sizes, profit-share pool sizing, small-business exit multiples — reflect market conditions as of July 2026 and move with the rate environment and the funding market. The mechanics are durable; the numbers are not. Check current data before quoting a threshold, and name the vintage.\n\n## Core axioms\n\n**You can raise later; you cannot un-raise.** The first institutional dollar deletes \"good enough\" as an available outcome: a $250M fund holding 10% at exit needs a $2.5B result to return itself once, so the $80M acquisition that clears your mortgage rounds to zero on the board's math (`capital-valuation.md`). Name the fund's size against the honest ceiling of the business before the first meeting; if it can't cover, you are volunteering to be a write-off. The limit is matching the fund to the outcome: a $30M pre-seed fund returns itself on a $300M exit. Later has arrived when you can name a payback period you cannot fund from margin; it hasn't when you can't say what the money buys.\n\n**The growth plan in the deck is one you'd never run with your own money.** Three shapes take venture money badly. Services-inflected businesses get funded on a software multiple and spend three years starving the delivery motion that was working. Slow compounders with long sales cycles or regulated buyers are excellent on a ten-year clock and unfundable on a fund's. The real-but-small market is the most honest: a business that tops out at $40M *[bench 2026-07]* is a very good company and a bad venture asset, and the deck that clears diligence is the one with an inflated TAM. The limit is services-as-wedge, where the mismatch is deliberate and time-boxed.\n\n**Margin-funded growth is capped by the gap between paying for a customer and being paid.** The venture path pre-funds that gap with a round; here it is funded by last quarter's profit, whi"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, and crisis. Use whenever someone is wrestling with a startup, scale-up, or early-stage company decision — hiring or firing, picking a co-founder, whether to raise at all, bootstrapping and staying independent, non-dilutive capital, selling the company, moats and positioning, how AI changes the way a startup team works, sells, and hires, who stays accountable when an agent does the work, or any \"I'm a founder and I'm not sure what to do\" moment. Trigger without an explicit ask for \"advice\" — \"should I…\", \"is it normal that…\", \"we're thinking about…\", and \"how do experienced founders handle…\" are in scope, as is helping someone else with founder-stage decisions (advisor, coach, mentor, investor). Also trigger on \"who said X\" questions about the startup canon. Skill: founder-wisdom Owner: chris-graffagnino Summary: Surfaces hard-won axioms from experienced startup founders and operators across hiring, fundraising, product, sales, finance, governance, strategy, management, and crisis. Use whenever someone is wrestling with a startup, scale-up, or early-stage company decision — hiring or firing, picking a co-founder, whether to raise at all, bootstrapping and staying indepen","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":2480,"uniquenessScore":45,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-11T19:18:11.503Z","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":"2026-10-11T19:18:11.503Z","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":"2026-10-11T21:57:22.150Z","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"}]}}}