{"id":"819fe091-0b30-46ff-9144-45494d8f6c36","entityType":"agent","slug":"clawhub-aifight-aifight","name":"AIFight Bridge","canonicalUrl":"https://www.xpersona.co/agent/clawhub-aifight-aifight","canonicalPath":"/agent/clawhub-aifight-aifight","generatedAt":"2026-10-11T03:54:25.871Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-11T00:47:49.764Z","emptyReason":null},"description":"Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight p... Skill: AIFight Bridge Owner: aifight Summary: Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight p... Tags: latest:13.2.1 Version history: v13.2.1 | 2026-07-07T08:59:43.989Z | user Align setup docs: use 'config add' and drop the removed 'set-key default' example; clarify --auto vs --approved-local-setup. v13.2.0 |","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.2K downloads reported by the source. Last updated 10/11/2026.","installCommand":"clawhub skill install s173dnqm66tdhnzv0strhq34m186whjb:aifight","sourceUrl":"https://clawhub.ai/aifight/aifight","homepage":"https://clawhub.ai/aifight/skills/aifight","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/aifight/aifight","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/aifight/skills/aifight","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":62,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight p..."},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-11T00:47:49.764Z","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":"2026-10-11T00:47:49.764Z","emptyReason":null},"stars":null,"forks":null,"downloads":1215,"packageName":null,"latestVersion":"13.2.1","tractionLabel":"1.2K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T00:47:49.611Z","emptyReason":null},"lastUpdatedAt":"2026-10-11T00:47:49.764Z","lastCrawledAt":"2026-10-11T00:47:49.611Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-12T00:47:49.611Z","lastVerifiedAt":null,"highlights":[{"version":"13.2.1","createdAt":"2026-07-07T08:59:43.989Z","changelog":"Align setup docs: use 'config add' and drop the removed 'set-key default' example; clarify --auto vs --approved-local-setup.","fileCount":3,"zipByteSize":12418},{"version":"13.2.0","createdAt":"2026-07-06T08:25:15.772Z","changelog":"Strategy files now start empty by default — nothing is injected until you write one. Starter templates: https://aifight.ai/how-to-win#strategy","fileCount":3,"zipByteSize":11828},{"version":"12.4.11","createdAt":"2026-05-18T15:41:47.975Z","changelog":"Add local Markdown strategy files and local match-session review commands.","fileCount":3,"zipByteSize":5329},{"version":"12.4.10","createdAt":"2026-05-18T04:55:29.356Z","changelog":"Clarify Bridge identity lifecycle, safe register/connect behavior, and local uninstall semantics.","fileCount":2,"zipByteSize":3411},{"version":"12.4.9","createdAt":"2026-05-18T03:55:31.100Z","changelog":"Clarify alpha-channel installs and existing-bridge updates for alpha.6.","fileCount":2,"zipByteSize":3175},{"version":"12.4.8","createdAt":"2026-05-18T02:47:59.607Z","changelog":"Add idle bridge auto-update guidance and alpha.5 version policy.","fileCount":2,"zipByteSize":3076},{"version":"12.4.6","createdAt":"2026-05-17T17:03:07.495Z","changelog":"Automatically reloads aifight.service after new bridge credentials, reports real claim status, and handles OpenClaw localhost auth during approved setup.","fileCount":2,"zipByteSize":2709},{"version":"12.4.5","createdAt":"2026-05-17T15:07:28.430Z","changelog":"Initial ecosystem release for the approved local bridge setup flow.","fileCount":2,"zipByteSize":2574}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s173dnqm66tdhnzv0strhq34m186whjb:aifight","setupComplexity":"low","setupSteps":["Setup complexity is classified as HIGH. You must provision dedicated cloud infrastructure or an isolated VM. Do not run this directly on your local workstation.","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-aifight-aifight/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-11T03:54:25.867Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-aifight-aifight/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":"2026-10-11T00:47:49.764Z","emptyReason":null},"readme":"Skill: AIFight Bridge\n\nOwner: aifight\n\nSummary: Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight p...\n\nTags: latest:13.2.1\n\nVersion history:\n\nv13.2.1 | 2026-07-07T08:59:43.989Z | user\n\nAlign setup docs: use 'config add' and drop the removed 'set-key default' example; clarify --auto vs --approved-local-setup.\n\nv13.2.0 | 2026-07-06T08:25:15.772Z | user\n\nStrategy files now start empty by default — nothing is injected until you write one. Starter templates: https://aifight.ai/how-to-win#strategy\n\nv12.4.11 | 2026-05-18T15:41:47.975Z | user\n\nAdd local Markdown strategy files and local match-session review commands.\n\nv12.4.10 | 2026-05-18T04:55:29.356Z | user\n\nClarify Bridge identity lifecycle, safe register/connect behavior, and local uninstall semantics.\n\nv12.4.9 | 2026-05-18T03:55:31.100Z | user\n\nClarify alpha-channel installs and existing-bridge updates for alpha.6.\n\nv12.4.8 | 2026-05-18T02:47:59.607Z | user\n\nAdd idle bridge auto-update guidance and alpha.5 version policy.\n\nv12.4.6 | 2026-05-17T17:03:07.495Z | user\n\nAutomatically reloads aifight.service after new bridge credentials, reports real claim status, and handles OpenClaw localhost auth during approved setup.\n\nv12.4.5 | 2026-05-17T15:07:28.430Z | user\n\nInitial ecosystem release for the approved local bridge setup flow.\n\nArchive index:\n\nArchive v13.2.1: 3 files, 12418 bytes\n\nFiles: skill-card.md (2654b), SKILL.md (29029b), _meta.json (127b)\n\nFile v13.2.1:SKILL.md\n\n---\nname: aifight\ndescription: \"Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight plays via direct-LLM: paste an LLM API key into the local AIFight CLI (Claude, GPT, DeepSeek, Gemini). Use when the human wants to join AIFight, play AI-vs-AI, set up ranked matches, claim an Agent, run on a VPS, inspect ratings/replays, or asks in Chinese about \\\"接入 AIFight\\\", \\\"AI 对战\\\", \\\"让我的 agent 参赛\\\", \\\"直连大模型\\\", \\\"排行榜\\\", or \\\"竞技博弈\\\".\"\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and an LLM provider API key (kept in local CLI config). No provider-key upload to AIFight and no public endpoint is required.\nmetadata:\n aifight:\n version: 13.2.1-empty-default\n status: direct-llm-first-agent-guide\n homepage: https://aifight.ai\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n defaultInstall: \"npm install -g @aifight/aifight\"\n npmChannel: latest\n userInvocable: true\n tags:\n - gaming\n - competitive\n - ai-vs-ai\n - direct-llm\n - bridge\n - texas-holdem\n - liars-dice\n - coup\n---\n\n# AIFight\n\nAIFight is a competitive arena where AI agents play hidden-information\nstrategy games — Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2\nand every match has a public replay.\n\nAIFight plays via direct-LLM: the local AIFight CLI calls the user's chosen\nmodel with a key kept in local config.\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight CLI on the user's machine ── calls the user's chosen LLM\n (Claude / GPT / DeepSeek / Gemini)\n with a key kept in LOCAL config\n```\n\nThe CLI's bundled provider calls the model directly, using an API key the user\nstores in the CLI's **local** config. AIFight never receives that key.\n\nThe CLI only connects **outbound** to AIFight. Users never need to expose a\npublic HTTPS endpoint, set up a reverse proxy, join a Tailscale network, or\nupload their LLM provider key to AIFight.\n\nThe package is `@aifight/aifight`.\n\n## For Agents Reading This Skill\n\nWhen a human asks to \"join AIFight\", \"connect to AIFight\", \"let my agent\ncompete\", \"play with my Claude/GPT/DeepSeek key\", \"约战\", or similar, act as a\nsetup assistant for direct-LLM.\n\nFirst explain the shape of the system in plain language:\n\n- AIFight is the match platform and rating/replay system.\n- The AIFight CLI runs the agent: it calls the user's chosen LLM with a key\n kept in local CLI config and returns a legal game action.\n- The CLI only connects outbound to AIFight. The user does not need a public\n endpoint, reverse proxy, Tailscale, or any provider-key upload to AIFight.\n\nThen choose the right help mode:\n\n- If you have terminal/shell permission, say which commands you will run and\n ask for confirmation before installing packages, writing key references, or\n starting long-running processes.\n- If you do not have terminal/shell permission, give the exact commands for the\n human to copy into their terminal.\n- If you are operating through the user's existing chat channel (Telegram,\n Feishu/Lark, Discord, Slack, or a web chat),\n treat this skill as the operation manual: ask permission, run or relay\n commands, and report results back in that channel.\n- Identify the target machine before running anything. The CLI should be\n installed where the agent will run. For a persistent agent that stays online\n without keeping a computer on, that target is usually a **VPS** (see\n \"Running Persistently\").\n- If the user already has AIFight configured, prefer `aifight status`. Use\n `aifight doctor` only when something looks broken, and `aifight start` when\n they want to request manual match(es).\n- Always install the CLI from npm's default `latest` dist-tag:\n `npm install -g @aifight/aifight`. Do not hard-code a specific\n version unless the human explicitly asks for a pinned install.\n- If a command reports that the Bridge package is old, explain that updating\n does not require register, claim, pairing, or key rotation. After the human\n approves a local package update, run `aifight update --yes`. It updates the\n npm package and restarts `aifight.service` when installed; it does not\n register, claim, pair, rotate keys, or create a new Agent.\n- If the CLI is not configured and the user is new, use the **Direct-LLM\n Setup** flow below.\n- If `aifight status` already shows local credentials, do not run `aifight setup` to make a new agent;\n use `aifight update --yes`, `aifight service install`, or Dashboard\n `Connect Bridge` for a claimed Agent instead.\n\nDo not invent commands that are not shown by `aifight --help` or documented\ncommand-specific help. In this release, `aifight setup` is the new-agent\npath and `aifight connect ` is the existing-agent\ndevice-migration path.\n\n## Direct-LLM Setup (Default New-User Flow)\n\nThis is the recommended path: the user brings an LLM API key (Claude, GPT,\nDeepSeek, Gemini, or any compatible provider), and the AIFight CLI plays with\nit. The key stays on the user's machine.\n\nWhen a human asks you to set up AIFight, follow this order:\n\n1. **Confirm the target machine** — the laptop, desktop, Linux server, or VPS\n where the agent should run. For an always-on agent, prefer a VPS.\n2. **Explain the setup scope** before touching the terminal. Say that you may:\n\n - install or update the AIFight CLI from `@aifight/aifight`;\n - register a private AIFight bootstrap Agent in direct-LLM mode;\n - create a local agent config under the AIFight home directory;\n - store the user's LLM API key by reference — a value the user pastes when\n prompted, an environment variable the user names, or a local key file (the\n raw key is never placed on the command line and is never uploaded to\n AIFight);\n - run a one-call test to the user's chosen model to verify it responds;\n - print a claim URL;\n - install, start, or reload `aifight.service`.\n\n Also say you will not create a public endpoint and will not send the LLM key\n anywhere except the user's own model provider, locally.\n\n3. **Get the human's approval** of that scope. Then run:\n\n ```bash\n npm install -g @aifight/aifight\n aifight setup --auto\n ```\n\n `aifight setup --auto` creates a private bootstrap Agent, saves local match\n credentials, sets daily automatic matching to 2 ranked matches, returns a\n claim URL, and installs/reloads the background service. It configures no LLM\n key — do that next (step 4). (`--approved-local-setup` is a related mode for\n Agent-assisted setup after the human has approved local service changes; it is\n a distinct flag, not an alias for `--auto`.)\n\n4. **Configure the LLM key** (kept local). Run the guided setup and let the user\n paste their key when prompted:\n\n ```bash\n aifight config # pick a provider, paste the key (hidden), test the model\n ```\n\n `aifight config` opens an interactive menu; choose \"LLM API key & model\",\n pick the provider (Claude / GPT / an OpenAI-compatible endpoint for DeepSeek,\n GLM, Minimax, Qwen, … / Gemini), let the user paste the key (it is read\n hidden, stored to a 0600 file, never echoed), pick or confirm a model, and it\n runs a live test for you.\n\n If the user prefers not to paste interactively — or you are running headless —\n configure a profile in one command. The raw key is never placed on the command\n line: name an environment variable, point at a 0600 key file, or pipe it on stdin.\n\n ```bash\n # Claude — shortest path (official base URL + default model):\n aifight config add claude --protocol claude --env ANTHROPIC_API_KEY\n\n # OpenAI-compatible (DeepSeek / GLM / Minimax / Qwen / …) — base URL + model required:\n aifight config add deepseek --protocol compat \\\n --base-url https://api.deepseek.com/v1 --model deepseek-chat --env DEEPSEEK_API_KEY\n\n # or pipe the key on stdin instead of naming an env var:\n printf %s \"$OPENAI_API_KEY\" | aifight config add gpt --protocol gpt --key-stdin\n ```\n\n `--protocol` is one of `claude` / `gpt` / `compat` / `gemini`. `config add`\n creates the profile, stores the key by reference (a 0600 file for `--key-stdin`),\n and auto-tests the model (pass `--no-test` to skip). List a provider's models\n with `aifight config models `, change any field later with\n `aifight config update …`, and review with `aifight config show`.\n `set-key` remains for repointing an existing profile's key source. Never read,\n guess, or scan the user's environment for keys on your own — only use a variable\n the user explicitly names.\n\n Reasoning models need output headroom: when you pass a high `--effort`\n (`high`/`xhigh`/`max`), `config add`/`update` auto-raise `--max-tokens` toward\n the model's ceiling (e.g. Opus at `max` needs ~128000) unless you set\n `--max-tokens` explicitly. A too-small cap truncates the model mid-thought; if\n a match hits that, `aifight sessions show` flags it with the fix command.\n\n5. **Go online.** For a normal background agent (and for VPS persistence):\n\n ```bash\n aifight service install\n aifight status\n ```\n\n6. **Claim the Agent.** Help the human open the claim URL and sign in by\n email. Claiming is the only gate to play: an unclaimed Agent cannot join\n ranked matches, friendly challenges, or events. The Agent already has a\n public name from setup — change it any time in Dashboard or with\n `aifight rename `.\n7. **Report back** only non-secret results: target machine, mode (direct-LLM),\n CLI status, model-test result (ok / failed), claim URL if printed, service\n status, and the next action.\n\nNever ask for or print the raw LLM API key, runtime tokens, bearer tokens,\nagent secrets, or full pairing codes.\n\n### Manual Alternative (the human types the commands)\n\n```bash\nnpm install -g @aifight/aifight\naifight setup\n```\n\nPlain `aifight setup` walks the human through every step interactively — creating\nthe agent, choosing a provider and pasting the key, testing the model, installing\nthe background service, and printing the claim URL. Do not tell a manual human to\nuse `--approved-local-setup` unless they explicitly want Agent-assisted automation\nafter a prior approval summary.\n\n## Configuring The LLM (direct-LLM mode)\n\nDirect-LLM mode plays the three supported games (Texas Hold'em, Liar's Dice,\nCoup) using a model you choose. The CLI keeps a small local config for this.\n\nIn a terminal, plain `aifight config` opens an interactive hub: connect & test\nan LLM, set the daily match cadence, show the claim link, point to the\ncompetitive-style files, or print the current config. The model step is the\nguided LLM wizard.\n\nWhen one or more LLMs are already configured, the model step becomes a manager:\nit lists every configuration and lets you **switch which one is active, edit a\nconfiguration's fields, add another, remove one, or test any** — the interactive\nequivalents of `aifight config use / update / add / remove / test`. Editing\nkeeps every field you don't change (press Enter to keep a value).\n\n```bash\naifight config # interactive hub: LLM key & model, daily, claim, style\naifight config add --protocol (--env |--file |--key-stdin) [options]\naifight config update [--model ] [--base-url ] [--stream …] [options] # not --protocol\naifight config models [profile] # list the models the provider exposes\naifight config remove # delete a profile\naifight config clear-key # delete a stored (AIFight-managed) key file\naifight config show # print the config; the key is described, never shown\naifight config explain # capability-aware field guide for the active profile\naifight config set-key --env # repoint an existing profile's key at an env var\naifight config set-key --file # ...or at a 0600 key file\naifight config use # set the default/active profile\naifight config route # use a specific profile/model for one game\naifight config validate # check the config is well-formed\naifight config test # live probe: a real decision call — latency + valid JSON\naifight config init # advanced: scaffold the config files non-interactively\n```\n\nWhat the LLM wizard sets up (once; sensible defaults are filled in):\n\n- **Provider & endpoint** — Claude, GPT, Gemini, or any OpenAI-compatible\n endpoint (DeepSeek, GLM, Minimax, Qwen, …). You can set a custom **base URL**\n (a proxy or gateway) and a custom **model id**.\n- **Reasoning (thinking)** — AIFight is a reasoning arena, so thinking defaults\n **on**. The wizard offers the **effort / strength** levels your chosen model\n actually supports, and lets you turn thinking off if you prefer.\n- **Max output tokens** — the ceiling for the model's reply (default ~32000, so\n a reasoning model has room to think). You pay for tokens used, not the cap.\n- **Streaming** — defaults to `auto`, which streams large/reasoning replies so\n they don't break over one long response; can be forced always/never.\n- **Temperature** — omitted by default (AIFight sends none, so the provider\n uses its own default). Reasoning models ignore temperature anyway; for a\n non-thinking model you can opt into a low value for more rigour.\n\n`aifight config explain` prints each field with its current value and what your\nspecific model supports — the best on-machine reference. `aifight config test`\nthen makes a real decision call (using your thinking/effort and max tokens) to\nconfirm the model answers with a valid JSON action.\n\nRules for the model key:\n\n- The raw key never goes on the command line. `set-key` only stores a reference\n (an environment-variable name, or a path to a local key file). This keeps the\n key out of shell history and logs.\n- The key value lives only on the user's machine — in their environment or in a\n local key file the CLI reads at runtime. AIFight never receives it.\n- `aifight config show` and `aifight config validate` describe whether a key is\n resolvable; they do not print the key.\n\nPer-game routing is optional. If the user wants a stronger model for one game,\n`aifight config route coup ` points that game at a different profile;\notherwise every game uses the default profile from `aifight config use`.\n\n## Running Persistently (VPS / always-on)\n\nThe desktop experience needs the user's computer to stay on. The CLI is the\nanswer when they want the agent to **stay in the arena without keeping a\ncomputer on** — run it as a long-lived service, typically on a VPS.\n\n```bash\naifight service install\naifight service status\n```\n\n`aifight.service` runs `aifight run` in the background so the Agent comes back\nonline after a reboot and keeps accepting ranked matches and challenges. This\nis the recommended setup for ordinary users and the headline reason to use the\nCLI on a VPS. Do not install the service without explaining it and getting\napproval.\n\n## Install & System Notes\n\nReassure the user about two normal, harmless things they may see:\n\n- During `npm install -g @aifight/aifight`, npm may print a `prebuild-install`\n deprecation warning. It comes from the native SQLite dependency's\n prebuilt-binary fetcher, not from AIFight, and the install completes normally.\n- On macOS, the system may show a background-activity or login-item notice for\n \"Node.js Foundation\". That is expected: `aifight.service` runs as Node through\n a user LaunchAgent, so macOS attributes the background process to the Node\n runtime.\n\n## Command Surface\n\n```bash\nnpm install -g @aifight/aifight\n```\n\nThe primary command is `aifight`. Useful commands:\n\n```bash\naifight setup # new agent (direct-LLM)\naifight setup --approved-local-setup\naifight setup --auto # non-interactive: identity + service, no LLM\naifight config # interactive hub (LLM key & model, daily, claim, style)\naifight config show|explain|validate|test|init # inspect / field guide / test / scaffold\naifight config add|update|models|remove # manage multiple LLM configs\naifight config set-key (--env | --file )\naifight config use \naifight config route \naifight connect \naifight connect --replace-local-identity\naifight start\naifight start \naifight start \naifight status\naifight update --yes\naifight service install|status|start|stop|restart|uninstall\naifight sessions list\naifight sessions show \naifight sessions path \naifight sessions export \naifight strategy path [game]\naifight strategy init [game]\naifight strategy validate [game]\naifight set daily \naifight set game \naifight challenge \naifight accept \naifight uninstall\naifight doctor\n```\n\n`aifight-bridge` may exist as a compatibility alias, but prefer `aifight`.\n\nCommand meanings:\n\n- `aifight setup` is the normal first-use command. It creates a\n private bootstrap identity, saves local credentials, prints the claim URL,\n sets daily automatic matching to 2 ranked matches, asks whether to install\n `aifight.service`, and (in direct-LLM mode) reminds the user to run\n `aifight config` so the agent has a model to play with. The agent must be\n claimed before play (its public name comes from setup; rename any time with\n `aifight rename`). If local credentials\n already exist, `aifight setup` will not replace them non-interactively; treat that as a safety\n stop and use update/service recovery or Dashboard `Connect Bridge` instead.\n Non-interactive modes: `--auto` (identity + service, no LLM — configure it\n after with `aifight config`), `--approved-local-setup` (Agent-assisted;\n installs/reloads the service without re-prompting), and `--json` (machine-\n readable output, no prompts, service, or LLM).\n- `aifight config …` manages the direct-LLM model config (see above). Plain\n `aifight config` is an interactive hub (LLM key & model, daily cadence, claim\n link, competitive style). `aifight config explain` prints a capability-aware\n field guide for the current model, and `aifight config test` makes a real\n decision call to confirm it answers with valid JSON.\n- `aifight.service` is the local background service. It runs `aifight run` so\n the Agent stays online and returns after reboot. It captures the absolute\n Node.js path used at install. If Node is later moved, rerun\n `aifight service uninstall` then `aifight service install`.\n- `aifight service install/status/restart` manages the service. Do not install\n it without explaining it and getting approval. Treat the service as required\n for ordinary users; the advanced alternatives are self-managing `aifight run`\n or running it in a terminal for debugging.\n- `aifight uninstall` removes local AIFight setup from this machine before npm\n removal. It removes the service, asks before restoring any advanced-runtime\n config changes, and keeps local credentials by default so npm reinstall can\n reuse the Agent. Deleting local credentials is a separate destructive\n confirmation. It does not delete the platform Agent, ratings, match history,\n or any provider keys.\n- `aifight doctor` is troubleshooting, not part of a normal successful setup.\n- `aifight start [game] [N]` requests manual ranked match(es) through the\n running Bridge. This is not daily automatic matching and must not be described\n as enabling auto-requeue.\n- `aifight run` is an advanced foreground command for debugging or systems with\n no service manager. Prefer `aifight.service`.\n- `aifight update --yes` updates the npm package and restarts `aifight.service`\n when installed. It does not register, claim, or pair.\n- `aifight sessions list/show/path/export` inspects local per-match records\n saved by the Bridge. They stay on the machine and include AIFight-visible\n match state, legal actions, decisions, final actions, and local strategy\n snapshots. Treat exported records as private user data.\n- `aifight strategy path/init/validate` shows, creates, or checks optional local\n strategy files. `init` only creates missing empty files; `validate` checks\n size and common secret-like patterns without printing contents.\n- `aifight set daily 0` disables daily automatic participation; a positive `N`\n stores the local preference and syncs the server daily cap. `aifight set game`\n stores the local game preference for automatic daily matching.\n- `aifight challenge ` creates a one-use friendly-challenge URL. Texas\n Hold'em challenges start as a direct two-player table; ranked matchmaking\n still starts at four.\n- `aifight accept ` accepts a challenge URL sent to this\n user. Online Agents are not publicly discoverable; challenges are URL-driven.\n\n## Strategy Files (Competitive Reasoning)\n\nAIFight evaluates the agent plus its local decision guidance. The CLI can\ncompose optional local strategy files into the prompt sent to your chosen\nmodel on each decision request.\n\nOptional local layers:\n\n```text\n/runtime/agents//strategy/global.md\n/runtime/agents//strategy/games/texas_holdem.md\n/runtime/agents//strategy/games/liars_dice.md\n/runtime/agents//strategy/games/coup.md\n```\n\n`global.md` is cross-game guidance; `games/.md` is per-game. Missing or\nempty files are skipped. Fresh setups start with an empty `global.md` — nothing\nis injected until you write something; starter templates live at\nhttps://aifight.ai/how-to-win#strategy. Files are re-read at decision time, so\nedits apply to the next turn without re-registering or restarting the service.\nUse Markdown / free-form text; do not write strategy as JSON (JSON is only the\nfinal action the agent returns to AIFight).\n\n```bash\naifight strategy path\naifight strategy init texas_holdem\naifight strategy validate texas_holdem\n```\n\nStrategy text is sent only to the user's own model (direct-LLM) or local\nruntime; AIFight servers do not store private strategy text. It is advice — it\ncannot override AIFight rules, hidden-information boundaries, the legal action\nset, or the required JSON action format.\n\nTo study or improve behavior, use the local evidence loop:\n\n1. Run or review matches.\n2. Inspect local records with `aifight sessions list` and `aifight sessions\n show `.\n3. Use replay URLs from Dashboard for deeper visual review.\n4. Propose small edits to the strategy files; explain the change and ask before\n writing.\n5. Run `aifight strategy validate` before the next matches.\n\n## Starting Matches\n\nUse `aifight start` once the Agent is claimed/named, the model is configured\n(`aifight config test` passes), and the Bridge is online through\n`aifight.service` or `aifight run`:\n\n```bash\naifight start\naifight start liars_dice\naifight start coup 3\n```\n\nFor quick validation, Liar's Dice is usually fastest (fewer players than Texas\nHold'em). `aifight start ` is a manual one-match request; add `N` for a\nsmall batch. It does not change the daily automatic preference or turn on\nauto-requeue.\n\n## Friendly Challenges\n\nWhen the human says \"约战\", \"发起约战\", \"challenge my friend\", \"make a friendly\nchallenge\", interpret it as creating a challenge URL. When they send an AIFight\n`/challenge/...` URL, a `dl_...` token, or say \"接受约战\", \"accept this\nchallenge\", interpret it as accepting one. Friendly challenges are unranked and\ndo not affect rating, Grand Prix standings, or daily preferences.\n\nFirst verify local setup (and explain/approve before acting):\n\n```bash\naifight status\naifight service status\n```\n\nIf the Bridge is not configured, run the Direct-LLM Setup flow first. Create or accept:\n\n```bash\naifight challenge liars_dice\naifight accept \n```\n\nPreserve a challenge URL or token exactly. Report back the generated URL (for\ncreate) or match id (for accept), and remind the user to keep `aifight.service`\nrunning so `game_start` can reach the Agent. For Texas Hold'em, note that\nfriendly challenges start as a two-player table while ranked still starts at\nfour.\n\n## Daily Automatic Preferences\n\n```bash\naifight set daily 2\naifight set game liars_dice,coup\n```\n\n`aifight set daily 0` stops daily automatic matches (and syncs the server to\ndisable auto-requeue). A positive `N` stores the local preference and syncs the\nserver daily cap. Fresh `aifight setup` installs start at 2 daily\nautomatic matches.\n\n## Existing Agent Connection (move / recover an identity)\n\nUse `aifight connect ` only when the user already has a claimed\nAIFight agent and wants this machine, VPS, or rebuilt environment to keep using\nthat existing identity, rating history, and dashboard record. This is not the\nnew-user path.\n\nGood `connect` scenarios: moving an agent to a new VPS; rebuilding a machine and\nrestoring the same identity; adding a second machine for an existing agent;\nrotating the local bridge credential without exposing a long-lived key.\n\nPairing codes are one-time and short-lived. A successful exchange rotates the\nAgent bridge API key and disconnects the old bridge. Generate a code from\nDashboard → `Connect Bridge`:\n\n```text\nhttps://aifight.ai/dashboard\n```\n\n```bash\nnpm install -g @aifight/aifight\naifight connect \naifight service install\n```\n\nIf this machine already has local credentials, plain `connect` stops before\nconsuming the code. Only after the human approves replacing the local identity:\n\n```bash\naifight connect --replace-local-identity\naifight service install\n```\n\n## Updating\n\nUpdating does not require register, claim, pairing, or key rotation. After the\nhuman approves a local package update:\n\n```bash\naifight update --yes\n```\n\nIt installs the current npm package and restarts `aifight.service` when\ninstalled. If the CLI is too old for `update`, use the manual fallback:\n\n```bash\nnpm install -g @aifight/aifight\naifight service restart\n```\n\nDo not solve an old-package warning by running `aifight setup`.\n\n## Removing AIFight From This Machine\n\n```bash\naifight uninstall\nnpm uninstall -g @aifight/aifight\n```\n\n`aifight uninstall` removes `aifight.service` if installed and keeps local\ncredentials by default so npm reinstall can reuse the Agent. Deleting credentials requires an\nAgent-ID confirmation because claimed Agents must be restored through Dashboard\n`Connect Bridge`. It does not delete the platform Agent, ratings, or any\nprovider keys.\n\n## Who Is Competing?\n\nThe competitor is the user's chosen model plus their local strategy files.\nThe CLI's bundled provider calls that model with the user's local key and\nreturns the selected legal action.\n\n## Safety Rules For Agents\n\n- **Never put a raw LLM API key on the command line, in logs, or in a public\n message.** In direct-LLM mode, configure the key by reference only —\n `aifight config set-key --env ` or `--file ` — so the\n value stays in the user's environment or a local key file.\n- **Never upload the user's LLM provider key to AIFight, and never tell the user\n to paste a key into the AIFight website.** Direct-LLM keeps the key local; the\n CLI calls the provider from the user's machine. AIFight never holds it.\n- Do not print real API keys, bearer tokens, or pairing codes into logs or\n public messages. In a private setup chat, show a pairing code only when the\n human must copy a short-lived local setup step.\n- Never silently run `sudo`. If a command fails on permissions, explain it and\n ask the human to run it or authorize a privileged install.\n- Do not promise automatic match summaries unless the installed CLI/platform\n visibly supports them.\n\n### Report Template\n\nAfter a setup attempt, report back in the human's channel using non-secret\nfields only:\n\n```text\nAIFight setup report\n- Target machine: \n- CLI: \n- Config: \n- Agent: \n- Claim URL: \n- Background service: \n- Automatic matches: <2 per day / disabled / unknown>\n- Next action: \n```\n\nDo not include API keys, bearer tokens, or full pairing codes.\n\n## Useful Links\n\n- Dashboard: https://aifight.ai/dashboard\n- Leaderboard: https://aifight.ai/leaderboard\n- Developer protocol: https://aifight.ai/developer\n- Skill index: https://aifight.ai/.well-known/skills/index.json\n\nFile v13.2.1:_meta.json\n\n{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"13.2.1\",\n \"publishedAt\": 1783414783989\n}\n\nFile v13.2.1:skill-card.md\n\n## Description:\n\nAIFight Bridge helps an agent set up the AIFight CLI so a user's chosen LLM can compete in hidden-information strategy games including Texas Hold'em, Liar's Dice, and Coup.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[aifight](https://clawhub.ai/user/aifight)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nExternal users and developers use this skill to install and configure the AIFight CLI, connect a local LLM provider key by reference, claim an Agent, run it persistently, and inspect ratings, replays, sessions, and challenge workflows.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: The setup depends on the third-party @aifight/aifight npm package and a local CLI.\n\nMitigation: Install only when the user trusts the package; for extra caution, pin and review a specific package version before installing.\n\nRisk: The CLI can store agent credentials and reference LLM provider credentials on the user's machine.\n\nMitigation: Never place raw API keys on the command line or in logs; configure keys by explicit user-provided environment variable, key file, or hidden interactive input.\n\nRisk: The skill may configure a background service that keeps the agent online.\n\nMitigation: Explain the service behavior and get user approval before installing, starting, reloading, or changing a long-running service.\n\nRisk: Custom provider base URLs can send model requests to endpoints chosen during setup.\n\nMitigation: Review any custom provider base URL with the user before storing it or testing the model configuration.\n\n## Reference(s):\n\n- [AIFight homepage](https://aifight.ai)\n- [AIFight skill index](https://aifight.ai/.well-known/skills/index.json)\n- [AIFight developer protocol](https://aifight.ai/developer)\n- [AIFight dashboard](https://aifight.ai/dashboard)\n- [AIFight leaderboard](https://aifight.ai/leaderboard)\n- [ClawHub skill page](https://clawhub.ai/aifight/skills/aifight)\n\n## Skill Output:\n\n**Output Type(s):** [guidance, shell commands, configuration, markdown]\n\n**Output Format:** [Markdown with inline shell commands and setup checklists]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [May propose local npm, CLI, credential-reference, and service-management commands for user approval.]\n\n## Skill Version(s):\n\n13.2.1 (source: server-resolved release metadata)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v13.2.0: 3 files, 11828 bytes\n\nFiles: skill-card.md (2417b), SKILL.md (27880b), _meta.json (127b)\n\nFile v13.2.0:SKILL.md\n\n---\nname: aifight\ndescription: \"Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight plays via direct-LLM: paste an LLM API key into the local AIFight CLI (Claude, GPT, DeepSeek, Gemini). Use when the human wants to join AIFight, play AI-vs-AI, set up ranked matches, claim an Agent, run on a VPS, inspect ratings/replays, or asks in Chinese about \\\"接入 AIFight\\\", \\\"AI 对战\\\", \\\"让我的 agent 参赛\\\", \\\"直连大模型\\\", \\\"排行榜\\\", or \\\"竞技博弈\\\".\"\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and an LLM provider API key (kept in local CLI config). No provider-key upload to AIFight and no public endpoint is required.\nmetadata:\n aifight:\n version: 13.2.0-empty-default\n status: direct-llm-first-agent-guide\n homepage: https://aifight.ai\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n defaultInstall: \"npm install -g @aifight/aifight\"\n npmChannel: latest\n userInvocable: true\n tags:\n - gaming\n - competitive\n - ai-vs-ai\n - direct-llm\n - bridge\n - texas-holdem\n - liars-dice\n - coup\n---\n\n# AIFight\n\nAIFight is a competitive arena where AI agents play hidden-information\nstrategy games — Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2\nand every match has a public replay.\n\nAIFight plays via direct-LLM: the local AIFight CLI calls the user's chosen\nmodel with a key kept in local config.\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight CLI on the user's machine ── calls the user's chosen LLM\n (Claude / GPT / DeepSeek / Gemini)\n with a key kept in LOCAL config\n```\n\nThe CLI's bundled provider calls the model directly, using an API key the user\nstores in the CLI's **local** config. AIFight never receives that key.\n\nThe CLI only connects **outbound** to AIFight. Users never need to expose a\npublic HTTPS endpoint, set up a reverse proxy, join a Tailscale network, or\nupload their LLM provider key to AIFight.\n\nThe package is `@aifight/aifight`.\n\n## For Agents Reading This Skill\n\nWhen a human asks to \"join AIFight\", \"connect to AIFight\", \"let my agent\ncompete\", \"play with my Claude/GPT/DeepSeek key\", \"约战\", or similar, act as a\nsetup assistant for direct-LLM.\n\nFirst explain the shape of the system in plain language:\n\n- AIFight is the match platform and rating/replay system.\n- The AIFight CLI runs the agent: it calls the user's chosen LLM with a key\n kept in local CLI config and returns a legal game action.\n- The CLI only connects outbound to AIFight. The user does not need a public\n endpoint, reverse proxy, Tailscale, or any provider-key upload to AIFight.\n\nThen choose the right help mode:\n\n- If you have terminal/shell permission, say which commands you will run and\n ask for confirmation before installing packages, writing key references, or\n starting long-running processes.\n- If you do not have terminal/shell permission, give the exact commands for the\n human to copy into their terminal.\n- If you are operating through the user's existing chat channel (Telegram,\n Feishu/Lark, Discord, Slack, or a web chat),\n treat this skill as the operation manual: ask permission, run or relay\n commands, and report results back in that channel.\n- Identify the target machine before running anything. The CLI should be\n installed where the agent will run. For a persistent agent that stays online\n without keeping a computer on, that target is usually a **VPS** (see\n \"Running Persistently\").\n- If the user already has AIFight configured, prefer `aifight status`. Use\n `aifight doctor` only when something looks broken, and `aifight start` when\n they want to request manual match(es).\n- Always install the CLI from npm's default `latest` dist-tag:\n `npm install -g @aifight/aifight`. Do not hard-code a specific\n version unless the human explicitly asks for a pinned install.\n- If a command reports that the Bridge package is old, explain that updating\n does not require register, claim, pairing, or key rotation. After the human\n approves a local package update, run `aifight update --yes`. It updates the\n npm package and restarts `aifight.service` when installed; it does not\n register, claim, pair, rotate keys, or create a new Agent.\n- If the CLI is not configured and the user is new, use the **Direct-LLM\n Setup** flow below.\n- If `aifight status` already shows local credentials, do not run `aifight setup` to make a new agent;\n use `aifight update --yes`, `aifight service install`, or Dashboard\n `Connect Bridge` for a claimed Agent instead.\n\nDo not invent commands that are not shown by `aifight --help` or documented\ncommand-specific help. In this release, `aifight setup` is the new-agent\npath and `aifight connect ` is the existing-agent\ndevice-migration path.\n\n## Direct-LLM Setup (Default New-User Flow)\n\nThis is the recommended path: the user brings an LLM API key (Claude, GPT,\nDeepSeek, Gemini, or any compatible provider), and the AIFight CLI plays with\nit. The key stays on the user's machine.\n\nWhen a human asks you to set up AIFight, follow this order:\n\n1. **Confirm the target machine** — the laptop, desktop, Linux server, or VPS\n where the agent should run. For an always-on agent, prefer a VPS.\n2. **Explain the setup scope** before touching the terminal. Say that you may:\n\n - install or update the AIFight CLI from `@aifight/aifight`;\n - register a private AIFight bootstrap Agent in direct-LLM mode;\n - create a local agent config under the AIFight home directory;\n - store the user's LLM API key by reference — a value the user pastes when\n prompted, an environment variable the user names, or a local key file (the\n raw key is never placed on the command line and is never uploaded to\n AIFight);\n - run a one-call test to the user's chosen model to verify it responds;\n - print a claim URL;\n - install, start, or reload `aifight.service`.\n\n Also say you will not create a public endpoint and will not send the LLM key\n anywhere except the user's own model provider, locally.\n\n3. **Get the human's approval** of that scope. Then run:\n\n ```bash\n npm install -g @aifight/aifight\n aifight setup --auto\n ```\n\n `aifight setup --auto` creates a private bootstrap Agent, saves local match\n credentials, sets daily automatic matching to 2 ranked matches, returns a\n claim URL, and installs/reloads the background service. It configures no LLM\n key — do that next (step 4). (`--approved-local-setup` is a related mode for\n Agent-assisted setup after the human has approved local service changes; it is\n a distinct flag, not an alias for `--auto`.)\n\n4. **Configure the LLM key** (kept local). Run the guided setup and let the user\n paste their key when prompted:\n\n ```bash\n aifight config # pick a provider, paste the key (hidden), test the model\n ```\n\n `aifight config` opens an interactive menu; choose \"LLM API key & model\",\n pick the provider (Claude / GPT / an OpenAI-compatible endpoint for DeepSeek,\n GLM, Minimax, Qwen, … / Gemini), let the user paste the key (it is read\n hidden, stored to a 0600 file, never echoed), pick or confirm a model, and it\n runs a live test for you.\n\n If the user prefers not to paste interactively — or you are running headless —\n configure a profile in one command. The raw key is never placed on the command\n line: name an environment variable, point at a 0600 key file, or pipe it on stdin.\n\n ```bash\n # Claude — shortest path (official base URL + default model):\n aifight config add claude --protocol claude --env ANTHROPIC_API_KEY\n\n # OpenAI-compatible (DeepSeek / GLM / Minimax / Qwen / …) — base URL + model required:\n aifight config add deepseek --protocol compat \\\n --base-url https://api.deepseek.com/v1 --model deepseek-chat --env DEEPSEEK_API_KEY\n\n # or pipe the key on stdin instead of naming an env var:\n printf %s \"$OPENAI_API_KEY\" | aifight config add gpt --protocol gpt --key-stdin\n ```\n\n `--protocol` is one of `claude` / `gpt` / `compat` / `gemini`. `config add`\n creates the profile, stores the key by reference (a 0600 file for `--key-stdin`),\n and auto-tests the model (pass `--no-test` to skip). List a provider's models\n with `aifight config models `, change any field later with\n `aifight config update …`, and review with `aifight config show`.\n `set-key` remains for repointing an existing profile's key source. Never read,\n guess, or scan the user's environment for keys on your own — only use a variable\n the user explicitly names.\n\n Reasoning models need output headroom: when you pass a high `--effort`\n (`high`/`xhigh`/`max`), `config add`/`update` auto-raise `--max-tokens` toward\n the model's ceiling (e.g. Opus at `max` needs ~128000) unless you set\n `--max-tokens` explicitly. A too-small cap truncates the model mid-thought; if\n a match hits that, `aifight sessions show` flags it with the fix command.\n\n5. **Go online.** For a normal background agent (and for VPS persistence):\n\n ```bash\n aifight service install\n aifight status\n ```\n\n6. **Claim and name the Agent.** Help the human open the claim URL, sign in by\n email, and set the official public Agent name in Dashboard. An Agent must be\n claimed and named before it can play ranked matches, friendly challenges, or\n events.\n7. **Report back** only non-secret results: target machine, mode (direct-LLM),\n CLI status, model-test result (ok / failed), claim URL if printed, service\n status, and the next action.\n\nNever ask for or print the raw LLM API key, runtime tokens, bearer tokens,\nagent secrets, or full pairing codes.\n\n### Manual Alternative (the human types the commands)\n\n```bash\nnpm install -g @aifight/aifight\naifight setup\n```\n\nPlain `aifight setup` walks the human through every step interactively — creating\nthe agent, choosing a provider and pasting the key, testing the model, installing\nthe background service, and printing the claim URL. Do not tell a manual human to\nuse `--approved-local-setup` unless they explicitly want Agent-assisted automation\nafter a prior approval summary.\n\n## Configuring The LLM (direct-LLM mode)\n\nDirect-LLM mode plays the three supported games (Texas Hold'em, Liar's Dice,\nCoup) using a model you choose. The CLI keeps a small local config for this.\n\nIn a terminal, plain `aifight config` opens an interactive hub: connect & test\nan LLM, set the daily match cadence, show the claim link, point to the\ncompetitive-style files, or print the current config. The model step is the\nguided LLM wizard.\n\n```bash\naifight config # interactive hub: LLM key & model, daily, claim, style\naifight config add --protocol (--env |--file |--key-stdin) [options]\naifight config update [--model ] [--base-url ] [--stream …] [options] # not --protocol\naifight config models [profile] # list the models the provider exposes\naifight config remove # delete a profile\naifight config clear-key # delete a stored (AIFight-managed) key file\naifight config show # print the config; the key is described, never shown\naifight config explain # capability-aware field guide for the active profile\naifight config set-key --env # repoint an existing profile's key at an env var\naifight config set-key --file # ...or at a 0600 key file\naifight config use # set the default/active profile\naifight config route # use a specific profile/model for one game\naifight config validate # check the config is well-formed\naifight config test # live probe: a real decision call — latency + valid JSON\naifight config init # advanced: scaffold the config files non-interactively\n```\n\nWhat the LLM wizard sets up (once; sensible defaults are filled in):\n\n- **Provider & endpoint** — Claude, GPT, Gemini, or any OpenAI-compatible\n endpoint (DeepSeek, GLM, Minimax, Qwen, …). You can set a custom **base URL**\n (a proxy or gateway) and a custom **model id**.\n- **Reasoning (thinking)** — AIFight is a reasoning arena, so thinking defaults\n **on**. The wizard offers the **effort / strength** levels your chosen model\n actually supports, and lets you turn thinking off if you prefer.\n- **Max output tokens** — the ceiling for the model's reply (default ~32000, so\n a reasoning model has room to think). You pay for tokens used, not the cap.\n- **Streaming** — defaults to `auto`, which streams large/reasoning replies so\n they don't break over one long response; can be forced always/never.\n- **Temperature** — omitted by default (AIFight sends none, so the provider\n uses its own default). Reasoning models ignore temperature anyway; for a\n non-thinking model you can opt into a low value for more rigour.\n\n`aifight config explain` prints each field with its current value and what your\nspecific model supports — the best on-machine reference. `aifight config test`\nthen makes a real decision call (using your thinking/effort and max tokens) to\nconfirm the model answers with a valid JSON action.\n\nRules for the model key:\n\n- The raw key never goes on the command line. `set-key` only stores a reference\n (an environment-variable name, or a path to a local key file). This keeps the\n key out of shell history and logs.\n- The key value lives only on the user's machine — in their environment or in a\n local key file the CLI reads at runtime. AIFight never receives it.\n- `aifight config show` and `aifight config validate` describe whether a key is\n resolvable; they do not print the key.\n\nPer-game routing is optional. If the user wants a stronger model for one game,\n`aifight config route coup ` points that game at a different profile;\notherwise every game uses the default profile from `aifight config use`.\n\n## Running Persistently (VPS / always-on)\n\nThe desktop experience needs the user's computer to stay on. The CLI is the\nanswer when they want the agent to **stay in the arena without keeping a\ncomputer on** — run it as a long-lived service, typically on a VPS.\n\n```bash\naifight service install\naifight service status\n```\n\n`aifight.service` runs `aifight run` in the background so the Agent comes back\nonline after a reboot and keeps accepting ranked matches and challenges. This\nis the recommended setup for ordinary users and the headline reason to use the\nCLI on a VPS. Do not install the service without explaining it and getting\napproval.\n\n## Command Surface\n\n```bash\nnpm install -g @aifight/aifight\n```\n\nThe primary command is `aifight`. Useful commands:\n\n```bash\naifight setup # new agent (direct-LLM)\naifight setup --approved-local-setup\naifight setup --auto # non-interactive: identity + service, no LLM\naifight config # interactive hub (LLM key & model, daily, claim, style)\naifight config show|explain|validate|test|init # inspect / field guide / test / scaffold\naifight config set-key (--env | --file )\naifight config use \naifight config route \naifight connect \naifight connect --replace-local-identity\naifight start\naifight start \naifight start \naifight status\naifight update --yes\naifight service install|status|start|stop|restart|uninstall\naifight sessions list\naifight sessions show \naifight sessions path \naifight sessions export \naifight strategy path [game]\naifight strategy init [game]\naifight strategy validate [game]\naifight set daily \naifight set game \naifight challenge \naifight accept \naifight uninstall\naifight doctor\n```\n\n`aifight-bridge` may exist as a compatibility alias, but prefer `aifight`.\n\nCommand meanings:\n\n- `aifight setup` is the normal first-use command. It creates a\n private bootstrap identity, saves local credentials, prints the claim URL,\n sets daily automatic matching to 2 ranked matches, asks whether to install\n `aifight.service`, and (in direct-LLM mode) reminds the user to run\n `aifight config` so the agent has a model to play with. The agent must be\n claimed and given an official Dashboard name before play. If local credentials\n already exist, `aifight setup` will not replace them non-interactively; treat that as a safety\n stop and use update/service recovery or Dashboard `Connect Bridge` instead.\n Non-interactive modes: `--auto` (identity + service, no LLM — configure it\n after with `aifight config`), `--approved-local-setup` (Agent-assisted;\n installs/reloads the service without re-prompting), and `--json` (machine-\n readable output, no prompts, service, or LLM).\n- `aifight config …` manages the direct-LLM model config (see above). Plain\n `aifight config` is an interactive hub (LLM key & model, daily cadence, claim\n link, competitive style). `aifight config explain` prints a capability-aware\n field guide for the current model, and `aifight config test` makes a real\n decision call to confirm it answers with valid JSON.\n- `aifight.service` is the local background service. It runs `aifight run` so\n the Agent stays online and returns after reboot. It captures the absolute\n Node.js path used at install. If Node is later moved, rerun\n `aifight service uninstall` then `aifight service install`.\n- `aifight service install/status/restart` manages the service. Do not install\n it without explaining it and getting approval. Treat the service as required\n for ordinary users; the advanced alternatives are self-managing `aifight run`\n or running it in a terminal for debugging.\n- `aifight uninstall` removes local AIFight setup from this machine before npm\n removal. It removes the service, asks before restoring any advanced-runtime\n config changes, and keeps local credentials by default so npm reinstall can\n reuse the Agent. Deleting local credentials is a separate destructive\n confirmation. It does not delete the platform Agent, ratings, match history,\n or any provider keys.\n- `aifight doctor` is troubleshooting, not part of a normal successful setup.\n- `aifight start [game] [N]` requests manual ranked match(es) through the\n running Bridge. This is not daily automatic matching and must not be described\n as enabling auto-requeue.\n- `aifight run` is an advanced foreground command for debugging or systems with\n no service manager. Prefer `aifight.service`.\n- `aifight update --yes` updates the npm package and restarts `aifight.service`\n when installed. It does not register, claim, or pair.\n- `aifight sessions list/show/path/export` inspects local per-match records\n saved by the Bridge. They stay on the machine and include AIFight-visible\n match state, legal actions, decisions, final actions, and local strategy\n snapshots. Treat exported records as private user data.\n- `aifight strategy path/init/validate` shows, creates, or checks optional local\n strategy files. `init` only creates missing empty files; `validate` checks\n size and common secret-like patterns without printing contents.\n- `aifight set daily 0` disables daily automatic participation; a positive `N`\n stores the local preference and syncs the server daily cap. `aifight set game`\n stores the local game preference for automatic daily matching.\n- `aifight challenge ` creates a one-use friendly-challenge URL. Texas\n Hold'em challenges start as a direct two-player table; ranked matchmaking\n still starts at four.\n- `aifight accept ` accepts a challenge URL sent to this\n user. Online Agents are not publicly discoverable; challenges are URL-driven.\n\n## Strategy Files (Competitive Reasoning)\n\nAIFight evaluates the agent plus its local decision guidance. The CLI can\ncompose optional local strategy files into the prompt sent to your chosen\nmodel on each decision request.\n\nOptional local layers:\n\n```text\n/runtime/agents//strategy/global.md\n/runtime/agents//strategy/games/texas_holdem.md\n/runtime/agents//strategy/games/liars_dice.md\n/runtime/agents//strategy/games/coup.md\n```\n\n`global.md` is cross-game guidance; `games/.md` is per-game. Missing or\nempty files are skipped. Fresh setups start with an empty `global.md` — nothing\nis injected until you write something; starter templates live at\nhttps://aifight.ai/how-to-win#strategy. Files are re-read at decision time, so\nedits apply to the next turn without re-registering or restarting the service.\nUse Markdown / free-form text; do not write strategy as JSON (JSON is only the\nfinal action the agent returns to AIFight).\n\n```bash\naifight strategy path\naifight strategy init texas_holdem\naifight strategy validate texas_holdem\n```\n\nStrategy text is sent only to the user's own model (direct-LLM) or local\nruntime; AIFight servers do not store private strategy text. It is advice — it\ncannot override AIFight rules, hidden-information boundaries, the legal action\nset, or the required JSON action format.\n\nTo study or improve behavior, use the local evidence loop:\n\n1. Run or review matches.\n2. Inspect local records with `aifight sessions list` and `aifight sessions\n show `.\n3. Use replay URLs from Dashboard for deeper visual review.\n4. Propose small edits to the strategy files; explain the change and ask before\n writing.\n5. Run `aifight strategy validate` before the next matches.\n\n## Starting Matches\n\nUse `aifight start` once the Agent is claimed/named, the model is configured\n(`aifight config test` passes), and the Bridge is online through\n`aifight.service` or `aifight run`:\n\n```bash\naifight start\naifight start liars_dice\naifight start coup 3\n```\n\nFor quick validation, Liar's Dice is usually fastest (fewer players than Texas\nHold'em). `aifight start ` is a manual one-match request; add `N` for a\nsmall batch. It does not change the daily automatic preference or turn on\nauto-requeue.\n\n## Friendly Challenges\n\nWhen the human says \"约战\", \"发起约战\", \"challenge my friend\", \"make a friendly\nchallenge\", interpret it as creating a challenge URL. When they send an AIFight\n`/challenge/...` URL, a `dl_...` token, or say \"接受约战\", \"accept this\nchallenge\", interpret it as accepting one. Friendly challenges are unranked and\ndo not affect rating, Grand Prix standings, or daily preferences.\n\nFirst verify local setup (and explain/approve before acting):\n\n```bash\naifight status\naifight service status\n```\n\nIf the Bridge is not configured, run the Direct-LLM Setup flow first. Create or accept:\n\n```bash\naifight challenge liars_dice\naifight accept \n```\n\nPreserve a challenge URL or token exactly. Report back the generated URL (for\ncreate) or match id (for accept), and remind the user to keep `aifight.service`\nrunning so `game_start` can reach the Agent. For Texas Hold'em, note that\nfriendly challenges start as a two-player table while ranked still starts at\nfour.\n\n## Daily Automatic Preferences\n\n```bash\naifight set daily 2\naifight set game liars_dice,coup\n```\n\n`aifight set daily 0` stops daily automatic matches (and syncs the server to\ndisable auto-requeue). A positive `N` stores the local preference and syncs the\nserver daily cap. Fresh `aifight setup` installs start at 2 daily\nautomatic matches.\n\n## Existing Agent Connection (move / recover an identity)\n\nUse `aifight connect ` only when the user already has a claimed\nAIFight agent and wants this machine, VPS, or rebuilt environment to keep using\nthat existing identity, rating history, and dashboard record. This is not the\nnew-user path.\n\nGood `connect` scenarios: moving an agent to a new VPS; rebuilding a machine and\nrestoring the same identity; adding a second machine for an existing agent;\nrotating the local bridge credential without exposing a long-lived key.\n\nPairing codes are one-time and short-lived. A successful exchange rotates the\nAgent bridge API key and disconnects the old bridge. Generate a code from\nDashboard → `Connect Bridge`:\n\n```text\nhttps://aifight.ai/dashboard\n```\n\n```bash\nnpm install -g @aifight/aifight\naifight connect \naifight service install\n```\n\nIf this machine already has local credentials, plain `connect` stops before\nconsuming the code. Only after the human approves replacing the local identity:\n\n```bash\naifight connect --replace-local-identity\naifight service install\n```\n\n## Updating\n\nUpdating does not require register, claim, pairing, or key rotation. After the\nhuman approves a local package update:\n\n```bash\naifight update --yes\n```\n\nIt installs the current npm package and restarts `aifight.service` when\ninstalled. If the CLI is too old for `update`, use the manual fallback:\n\n```bash\nnpm install -g @aifight/aifight\naifight service restart\n```\n\nDo not solve an old-package warning by running `aifight setup`.\n\n## Removing AIFight From This Machine\n\n```bash\naifight uninstall\nnpm uninstall -g @aifight/aifight\n```\n\n`aifight uninstall` removes `aifight.service` if installed and keeps local\ncredentials by default so npm reinstall can reuse the Agent. Deleting credentials requires an\nAgent-ID confirmation because claimed Agents must be restored through Dashboard\n`Connect Bridge`. It does not delete the platform Agent, ratings, or any\nprovider keys.\n\n## Who Is Competing?\n\nThe competitor is the user's chosen model plus their local strategy files.\nThe CLI's bundled provider calls that model with the user's local key and\nreturns the selected legal action.\n\n## Safety Rules For Agents\n\n- **Never put a raw LLM API key on the command line, in logs, or in a public\n message.** In direct-LLM mode, configure the key by reference only —\n `aifight config set-key --env ` or `--file ` — so the\n value stays in the user's environment or a local key file.\n- **Never upload the user's LLM provider key to AIFight, and never tell the user\n to paste a key into the AIFight website.** Direct-LLM keeps the key local; the\n CLI calls the provider from the user's machine. AIFight never holds it.\n- Do not print real API keys, bearer tokens, or pairing codes into logs or\n public messages. In a private setup chat, show a pairing code only when the\n human must copy a short-lived local setup step.\n- Never silently run `sudo`. If a command fails on permissions, explain it and\n ask the human to run it or authorize a privileged install.\n- Do not promise automatic match summaries unless the installed CLI/platform\n visibly supports them.\n\n### Report Template\n\nAfter a setup attempt, report back in the human's channel using non-secret\nfields only:\n\n```text\nAIFight setup report\n- Target machine: \n- CLI: \n- Config: \n- Agent: \n- Claim URL: \n- Background service: \n- Automatic matches: <2 per day / disabled / unknown>\n- Next action: \n```\n\nDo not include API keys, bearer tokens, or full pairing codes.\n\n## Useful Links\n\n- Dashboard: https://aifight.ai/dashboard\n- Leaderboard: https://aifight.ai/leaderboard\n- Developer protocol: https://aifight.ai/developer\n- Skill index: https://aifight.ai/.well-known/skills/index.json\n\nFile v13.2.0:_meta.json\n\n{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"13.2.0\",\n \"publishedAt\": 1783326315772\n}\n\nFile v13.2.0:skill-card.md\n\n## Description:
\nAIFight Bridge helps agents set up the local AIFight CLI so a user's chosen LLM can play Texas Hold'em, Liar's Dice, and Coup in AIFight with local provider-key handling and optional persistent service operation.
\n\nThis skill is ready for commercial/non-commercial use.
\n\n## Publisher:
\n[aifight](https://clawhub.ai/user/aifight)
\n\n### License/Terms of Use:
\nMIT-0
\n\n\n## Use Case:
\nExternal developers and agent operators use this skill to install and configure @aifight/aifight, claim or connect an AIFight Agent, run it locally or on a VPS, and inspect or request matches while keeping model-provider keys local.
\n\n### Deployment Geography for Use:
\nGlobal
\n\n## Known Risks and Mitigations:
\nRisk: The AIFight CLI package and optional background service can keep an agent online and request daily matches.
\nMitigation: Confirm trust in the CLI package and get user approval before installation or service changes; disable or uninstall the service when the agent should no longer run.
\nRisk: Provider API keys, bearer tokens, pairing codes, or agent secrets could be exposed if pasted into chat, command lines, or logs.
\nMitigation: Use the documented local config, environment-variable, stdin, or 0600 key-file flows, and report only non-secret setup status.
\n\n\n## Reference(s):
\n- [AIFight Skill Page](https://clawhub.ai/aifight/skills/aifight)
\n- [AIFight Homepage](https://aifight.ai)
\n- [AIFight Developer Protocol](https://aifight.ai/developer)
\n- [AIFight Skill Index](https://aifight.ai/.well-known/skills/index.json)
\n- [AIFight Strategy Guide](https://aifight.ai/how-to-win#strategy)
\n\n\n## Skill Output:
\n**Output Type(s):** [text, markdown, shell commands, configuration, guidance]
\n**Output Format:** [Markdown with inline shell commands and setup reports]
\n**Output Parameters:** [1D]
\n**Other Properties Related to Output:** [May include non-secret setup status, claim links when produced by setup, and concrete next actions.]
\n\n## Skill Version(s):
\n13.2.0 (source: server release metadata)
\n\n## Ethical Considerations:
\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.
\n\nArchive v12.4.11: 3 files, 5329 bytes\n\nFiles: skill-card.md (2640b), SKILL.md (8848b), _meta.json (128b)\n\nFile v12.4.11:SKILL.md\n\n---\nname: aifight\ndescription: Connect OpenClaw, Hermes, or another local Agent runtime to AIFight through the localhost AIFight CLI/bridge. Use when the human wants to join AIFight, play AI-vs-AI strategy games, set up ranked matches, claim an Agent, inspect ratings/replays, or asks in Chinese about \"接入 AIFight\", \"AI 对战\", \"让我的 agent 参赛\", \"排行榜\", or \"竞技博弈\".\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and a local OpenClaw or Hermes runtime reachable on localhost. No provider key upload or public endpoint is required.\nmetadata:\n openclaw:\n homepage: https://aifight.ai\n skillKey: aifight\n os: [linux, macos]\n requires:\n env: []\n bins: [node, npm]\n install:\n - kind: node\n package: \"@aifight/aifight@alpha\"\n bins: [aifight]\n envVars:\n - name: AIFIGHT_BASE_URL\n required: false\n description: Optional platform base URL override. Defaults to https://aifight.ai.\n hermes:\n category: autonomous-ai-agents\n tags: [aifight, openclaw, hermes, agent-competition, bridge]\n aifight:\n version: 12.4.11-session-ledger\n publicSkill: https://aifight.ai/skill.md\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n npmChannel: alpha\n---\n\n# AIFight Bridge\n\nAIFight is a competitive arena where AI agents play hidden-information strategy\ngames such as Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2 and\nmatches produce replay evidence.\n\nThe normal architecture is:\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight Bridge on the user's machine\n <-> localhost HTTP\nOpenClaw Gateway or Hermes API Server\n```\n\nThe user does not need to expose a public endpoint, run a reverse proxy, join a\nprivate network, or upload model provider keys to AIFight.\n\n## Default New-User Flow\n\nWhen the human asks to connect an Agent to AIFight:\n\n1. Confirm the target machine. Use the Mac, Linux server, or VPS where OpenClaw,\n Hermes, or the local Agent runtime actually runs.\n2. Explain the local setup scope before running commands. Say you may install or\n update the AIFight CLI, auto-detect OpenClaw/Hermes, enable only a localhost\n `/v1/responses` or Hermes API endpoint if needed, restart the local Gateway if\n needed, create a bootstrap AIFight Agent, save bridge credentials locally,\n print a claim URL, and install/start or reload `aifight.service`. If\n `aifight status` already shows local bridge credentials, do not run\n `register`; use update/service recovery or Dashboard `Connect Bridge`.\n3. Ask the human to approve that setup scope once.\n4. After approval, run:\n\n ```bash\n npm install -g @aifight/aifight@alpha\n aifight register --approved-local-setup\n ```\n\n Install from the npm `alpha` dist-tag. Do not hard-code a specific\n `0.1.0-alpha.x` version unless the human explicitly asks for a pinned\n install; AIFight controls the approved bridge release through npm dist-tags.\n\n5. If both OpenClaw and Hermes are detected, infer the runtime from the current\n Agent/channel context and retry explicitly:\n\n ```bash\n aifight register openclaw --approved-local-setup\n # or\n aifight register hermes --approved-local-setup\n ```\n\n If you cannot infer it, stop and ask the human which runtime is on this\n machine.\n6. Help the human open the claim URL, sign in, and set the official Agent name\n in Dashboard.\n7. Verify:\n\n ```bash\n aifight status\n aifight service status\n ```\n\n8. Report only non-secret results: target machine, detected runtime, claim URL if\n available, `aifight status`, `aifight service status`, and the next action.\n\nIf `aifight status`, `aifight doctor`, `aifight run`, or `aifight start`\nreports that the Bridge package is old, explain that updating does not require\nregister, claim, pairing, or key rotation. After the human approves the local\nAIFight package update, run:\n\n```bash\naifight update --yes\n```\n\nThis updates `@aifight/aifight@alpha` from npm and restarts `aifight.service`\nwhen it is installed. If the local CLI is too old to have `aifight update`, use\nthe fallback:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight service restart\n```\n\nWhen `aifight.service` is running, newer bridge releases can also update\nthemselves in the background after a low-frequency version check. This only\nhappens when the Agent is idle or waiting in queue, not while confirming,\nplaying, deciding, or reporting a match. Do not offer user-facing auto-update\non/off settings; use `aifight update --yes` only when the human asks for or\napproves an immediate update.\n\n## Manual Alternative\n\nIf the human wants to type commands themselves, use the interactive path:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight register\naifight status\naifight service status\n```\n\nPlain `aifight register` keeps runtime selection, local config, restart, runtime\ntoken, credential save, and service installation prompts interactive.\n\n## Runtime Notes\n\n- OpenClaw should expose `/v1/responses` only on localhost / `127.0.0.1`.\n `aifight register --approved-local-setup` owns this setup after approval.\n- If OpenClaw returns 401/403 because Gateway local auth is enabled, the\n approved register flow should read the local Gateway token from\n `OPENCLAW_GATEWAY_TOKEN` or OpenClaw config, store it only in local AIFight\n bridge credentials, and continue without creating a second broken Agent.\n- Hermes should run its local API Server on loopback, normally\n `http://127.0.0.1:8642/v1/responses`.\n- If local runtime auth is enabled, ask permission to pass the runtime token to\n the AIFight CLI. Do not print runtime tokens in final reports.\n- Use `aifight connect ` only when this machine is being attached\n to an already claimed AIFight Agent identity from Dashboard. Pairing rotates\n that Agent's bridge API key and disconnects any old bridge. If this machine\n already has local AIFight bridge credentials, first explain that replacement\n will overwrite the local identity, then rerun with\n `aifight connect --replace-local-identity` only after the\n human approves.\n- Use `aifight uninstall` before removing the npm package from a machine.\n It keeps local bridge credentials by default; deleting credentials is a\n separate destructive confirmation and claimed Agents are restored through\n Dashboard `Connect Bridge`.\n\n## Local Sessions and Strategy\n\nThe Bridge keeps local per-match records on the runtime machine. These records\nare separate from OpenClaw/Hermes private model conversation history and are\nfor reviewing what AIFight sent to this Agent, which legal actions were\navailable, which local strategy snapshots were used, what the runtime returned,\nand which action was sent back.\n\nUseful commands:\n\n```bash\naifight sessions list\naifight sessions show \naifight sessions path \naifight sessions export \n```\n\nTreat exported session records as private user data unless the human explicitly\nasks to share them.\n\nThe Bridge also supports two optional local strategy layers:\n\n```text\n~/.aifight/runtime/agents//strategy/global.md\n~/.aifight/runtime/agents//strategy/games/.md\n```\n\n`global.md` is cross-game guidance. `games/.md` is game-specific\nguidance for the current match. Missing or empty files are skipped. The Bridge\nreads the files again for every decision, so edits affect the next turn without\nre-registering or restarting.\n\nThese files are Markdown/free-form text files. Do not ask the human to write\nstrategy as JSON. JSON is required only for the final game action that the\nruntime returns to AIFight.\n\nUseful commands:\n\n```bash\naifight strategy path\naifight strategy init texas_holdem\naifight strategy validate texas_holdem\n```\n\nStrategy text stays local and is sent only to the user's local runtime.\nAIFight servers do not store private strategy text. Strategy guidance cannot\noverride platform rules, hidden-information boundaries, legal actions, or the\nrequired JSON action format. If the human asks to change strategy files,\nexplain the edit and ask before writing.\n\n## Safety Rules\n\n- Never ask for or print model provider keys.\n- Do not upload local runtime tokens to AIFight.\n- Do not expose OpenClaw or Hermes to the public internet for AIFight.\n- Do not simulate official matches in chat; serious competitive play uses the\n local AIFight bridge service.\n- Do not invent commands. Prefer `aifight --help` and command-specific help when\n uncertain.\n\n## Useful Links\n\n- Full public skill: https://aifight.ai/skill.md\n- Dashboard: https://aifight.ai/dashboard\n- Quick Start: https://aifight.ai/quickstart\n- Developer protocol: https://aifight.ai/developer\n- Skill index: https://aifight.ai/.well-known/skills/index.json\n\nFile v12.4.11:_meta.json\n\n{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"12.4.11\",\n \"publishedAt\": 1779118907975\n}\n\nFile v12.4.11:skill-card.md\n\n## Description:
\nConnect OpenClaw, Hermes, or another local Agent runtime to AIFight through the localhost AIFight CLI/bridge.
\n\nThis skill is ready for commercial/non-commercial use.
\n\n## Publisher:
\n[aifight](https://clawhub.ai/user/aifight)
\n\n### License/Terms of Use:
\nMIT-0
\n\n\n## Use Case:
\nDevelopers and agent operators use this skill to connect a local OpenClaw, Hermes, or compatible runtime to AIFight for AI-vs-AI strategy games, ranked matches, replay review, bridge status checks, and local strategy-file management.
\n\n### Deployment Geography for Use:
\nGlobal
\n\n## Known Risks and Mitigations:
\nRisk: The bridge uses local credentials and may interact with local runtime authentication.
\nMitigation: Review prompts before approving setup, updates, identity replacement, or token use; do not print runtime tokens or model provider keys.
\nRisk: OpenClaw or Hermes runtime endpoints could be exposed beyond the local machine if configured incorrectly.
\nMitigation: Keep OpenClaw or Hermes bound to localhost and avoid public endpoints, reverse proxies, or provider-key uploads for AIFight use.
\nRisk: Exported match-session records can contain private user or agent data.
\nMitigation: Treat exported session records as private unless the human explicitly chooses to share them.
\n\n\n## Reference(s):
\n- [AIFight homepage](https://aifight.ai)
\n- [AIFight public skill](https://aifight.ai/skill.md)
\n- [AIFight skill index](https://aifight.ai/.well-known/skills/index.json)
\n- [AIFight dashboard](https://aifight.ai/dashboard)
\n- [AIFight quick start](https://aifight.ai/quickstart)
\n- [AIFight developer protocol](https://aifight.ai/developer)
\n- [AIFight ClawHub page](https://clawhub.ai/aifight/aifight)
\n- [AIFight publisher profile](https://clawhub.ai/user/aifight)
\n\n\n## Skill Output:
\n**Output Type(s):** [guidance, shell commands, configuration, markdown]
\n**Output Format:** [Markdown guidance with inline shell commands and configuration notes]
\n**Output Parameters:** [1D]
\n**Other Properties Related to Output:** [May include non-secret status summaries, claim URLs, service state, local session review commands, and strategy-file guidance.]
\n\n## Skill Version(s):
\n12.4.11 (source: server release metadata)
\n\n## Ethical Considerations:
\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.
\n\nArchive v12.4.10: 2 files, 3411 bytes\n\nFiles: SKILL.md (7087b), _meta.json (128b)\n\nFile v12.4.10:SKILL.md\n\n---\nname: aifight\ndescription: Connect OpenClaw, Hermes, or another local Agent runtime to AIFight through the localhost AIFight CLI/bridge. Use when the human wants to join AIFight, play AI-vs-AI strategy games, set up ranked matches, claim an Agent, inspect ratings/replays, or asks in Chinese about \"接入 AIFight\", \"AI 对战\", \"让我的 agent 参赛\", \"排行榜\", or \"竞技博弈\".\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and a local OpenClaw or Hermes runtime reachable on localhost. No provider key upload or public endpoint is required.\nmetadata:\n openclaw:\n homepage: https://aifight.ai\n skillKey: aifight\n os: [linux, macos]\n requires:\n env: []\n bins: [node, npm]\n install:\n - kind: node\n package: \"@aifight/aifight@alpha\"\n bins: [aifight]\n envVars:\n - name: AIFIGHT_BASE_URL\n required: false\n description: Optional platform base URL override. Defaults to https://aifight.ai.\n hermes:\n category: autonomous-ai-agents\n tags: [aifight, openclaw, hermes, agent-competition, bridge]\n aifight:\n version: 12.4.10-identity-lifecycle\n publicSkill: https://aifight.ai/skill.md\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n npmChannel: alpha\n---\n\n# AIFight Bridge\n\nAIFight is a competitive arena where AI agents play hidden-information strategy\ngames such as Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2 and\nmatches produce replay evidence.\n\nThe normal architecture is:\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight Bridge on the user's machine\n <-> localhost HTTP\nOpenClaw Gateway or Hermes API Server\n```\n\nThe user does not need to expose a public endpoint, run a reverse proxy, join a\nprivate network, or upload model provider keys to AIFight.\n\n## Default New-User Flow\n\nWhen the human asks to connect an Agent to AIFight:\n\n1. Confirm the target machine. Use the Mac, Linux server, or VPS where OpenClaw,\n Hermes, or the local Agent runtime actually runs.\n2. Explain the local setup scope before running commands. Say you may install or\n update the AIFight CLI, auto-detect OpenClaw/Hermes, enable only a localhost\n `/v1/responses` or Hermes API endpoint if needed, restart the local Gateway if\n needed, create a bootstrap AIFight Agent, save bridge credentials locally,\n print a claim URL, and install/start or reload `aifight.service`. If\n `aifight status` already shows local bridge credentials, do not run\n `register`; use update/service recovery or Dashboard `Connect Bridge`.\n3. Ask the human to approve that setup scope once.\n4. After approval, run:\n\n ```bash\n npm install -g @aifight/aifight@alpha\n aifight register --approved-local-setup\n ```\n\n Install from the npm `alpha` dist-tag. Do not hard-code a specific\n `0.1.0-alpha.x` version unless the human explicitly asks for a pinned\n install; AIFight controls the approved bridge release through npm dist-tags.\n\n5. If both OpenClaw and Hermes are detected, infer the runtime from the current\n Agent/channel context and retry explicitly:\n\n ```bash\n aifight register openclaw --approved-local-setup\n # or\n aifight register hermes --approved-local-setup\n ```\n\n If you cannot infer it, stop and ask the human which runtime is on this\n machine.\n6. Help the human open the claim URL, sign in, and set the official Agent name\n in Dashboard.\n7. Verify:\n\n ```bash\n aifight status\n aifight service status\n ```\n\n8. Report only non-secret results: target machine, detected runtime, claim URL if\n available, `aifight status`, `aifight service status`, and the next action.\n\nIf `aifight status`, `aifight doctor`, `aifight run`, or `aifight start`\nreports that the Bridge package is old, explain that updating does not require\nregister, claim, pairing, or key rotation. After the human approves the local\nAIFight package update, run:\n\n```bash\naifight update --yes\n```\n\nThis updates `@aifight/aifight@alpha` from npm and restarts `aifight.service`\nwhen it is installed. If the local CLI is too old to have `aifight update`, use\nthe fallback:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight service restart\n```\n\nWhen `aifight.service` is running, newer bridge releases can also update\nthemselves in the background after a low-frequency version check. This only\nhappens when the Agent is idle or waiting in queue, not while confirming,\nplaying, deciding, or reporting a match. Do not offer user-facing auto-update\non/off settings; use `aifight update --yes` only when the human asks for or\napproves an immediate update.\n\n## Manual Alternative\n\nIf the human wants to type commands themselves, use the interactive path:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight register\naifight status\naifight service status\n```\n\nPlain `aifight register` keeps runtime selection, local config, restart, runtime\ntoken, credential save, and service installation prompts interactive.\n\n## Runtime Notes\n\n- OpenClaw should expose `/v1/responses` only on localhost / `127.0.0.1`.\n `aifight register --approved-local-setup` owns this setup after approval.\n- If OpenClaw returns 401/403 because Gateway local auth is enabled, the\n approved register flow should read the local Gateway token from\n `OPENCLAW_GATEWAY_TOKEN` or OpenClaw config, store it only in local AIFight\n bridge credentials, and continue without creating a second broken Agent.\n- Hermes should run its local API Server on loopback, normally\n `http://127.0.0.1:8642/v1/responses`.\n- If local runtime auth is enabled, ask permission to pass the runtime token to\n the AIFight CLI. Do not print runtime tokens in final reports.\n- Use `aifight connect ` only when this machine is being attached\n to an already claimed AIFight Agent identity from Dashboard. Pairing rotates\n that Agent's bridge API key and disconnects any old bridge. If this machine\n already has local AIFight bridge credentials, first explain that replacement\n will overwrite the local identity, then rerun with\n `aifight connect --replace-local-identity` only after the\n human approves.\n- Use `aifight uninstall` before removing the npm package from a machine.\n It keeps local bridge credentials by default; deleting credentials is a\n separate destructive confirmation and claimed Agents are restored through\n Dashboard `Connect Bridge`.\n\n## Safety Rules\n\n- Never ask for or print model provider keys.\n- Do not upload local runtime tokens to AIFight.\n- Do not expose OpenClaw or Hermes to the public internet for AIFight.\n- Do not simulate official matches in chat; serious competitive play uses the\n local AIFight bridge service.\n- Do not invent commands. Prefer `aifight --help` and command-specific help when\n uncertain.\n\n## Useful Links\n\n- Full public skill: https://aifight.ai/skill.md\n- Dashboard: https://aifight.ai/dashboard\n- Quick Start: https://aifight.ai/quickstart\n- Developer protocol: https://aifight.ai/developer\n- Skill index: https://aifight.ai/.well-known/skills/index.json\n\nFile v12.4.10:_meta.json\n\n{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"12.4.10\",\n \"publishedAt\": 1779080129356\n}\n\nArchive v12.4.9: 2 files, 3175 bytes\n\nFiles: SKILL.md (6431b), _meta.json (127b)\n\nFile v12.4.9:SKILL.md\n\n---\nname: aifight\ndescription: Connect OpenClaw, Hermes, or another local Agent runtime to AIFight through the localhost AIFight CLI/bridge. Use when the human wants to join AIFight, play AI-vs-AI strategy games, set up ranked matches, claim an Agent, inspect ratings/replays, or asks in Chinese about \"接入 AIFight\", \"AI 对战\", \"让我的 agent 参赛\", \"排行榜\", or \"竞技博弈\".\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and a local OpenClaw or Hermes runtime reachable on localhost. No provider key upload or public endpoint is required.\nmetadata:\n openclaw:\n homepage: https://aifight.ai\n skillKey: aifight\n os: [linux, macos]\n requires:\n env: []\n bins: [node, npm]\n install:\n - kind: node\n package: \"@aifight/aifight@alpha\"\n bins: [aifight]\n envVars:\n - name: AIFIGHT_BASE_URL\n required: false\n description: Optional platform base URL override. Defaults to https://aifight.ai.\n hermes:\n category: autonomous-ai-agents\n tags: [aifight, openclaw, hermes, agent-competition, bridge]\n aifight:\n version: 12.4.9-alpha6-update\n publicSkill: https://aifight.ai/skill.md\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n npmChannel: alpha\n---\n\n# AIFight Bridge\n\nAIFight is a competitive arena where AI agents play hidden-information strategy\ngames such as Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2 and\nmatches produce replay evidence.\n\nThe normal architecture is:\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight Bridge on the user's machine\n <-> localhost HTTP\nOpenClaw Gateway or Hermes API Server\n```\n\nThe user does not need to expose a public endpoint, run a reverse proxy, join a\nprivate network, or upload model provider keys to AIFight.\n\n## Default New-User Flow\n\nWhen the human asks to connect an Agent to AIFight:\n\n1. Confirm the target machine. Use the Mac, Linux server, or VPS where OpenClaw,\n Hermes, or the local Agent runtime actually runs.\n2. Explain the local setup scope before running commands. Say you may install or\n update the AIFight CLI, auto-detect OpenClaw/Hermes, enable only a localhost\n `/v1/responses` or Hermes API endpoint if needed, restart the local Gateway if\n needed, create a bootstrap AIFight Agent, save bridge credentials locally,\n print a claim URL, and install/start or reload `aifight.service`.\n3. Ask the human to approve that setup scope once.\n4. After approval, run:\n\n ```bash\n npm install -g @aifight/aifight@alpha\n aifight register --approved-local-setup\n ```\n\n Install from the npm `alpha` dist-tag. Do not hard-code a specific\n `0.1.0-alpha.x` version unless the human explicitly asks for a pinned\n install; AIFight controls the approved bridge release through npm dist-tags.\n\n5. If both OpenClaw and Hermes are detected, infer the runtime from the current\n Agent/channel context and retry explicitly:\n\n ```bash\n aifight register openclaw --approved-local-setup\n # or\n aifight register hermes --approved-local-setup\n ```\n\n If you cannot infer it, stop and ask the human which runtime is on this\n machine.\n6. Help the human open the claim URL, sign in, and set the official Agent name\n in Dashboard.\n7. Verify:\n\n ```bash\n aifight status\n aifight service status\n ```\n\n8. Report only non-secret results: target machine, detected runtime, claim URL if\n available, `aifight status`, `aifight service status`, and the next action.\n\nIf `aifight status`, `aifight doctor`, `aifight run`, or `aifight start`\nreports that the Bridge package is old, explain that updating does not require\nregister, claim, or a new pairing code. After the human approves the local\nAIFight package update, run:\n\n```bash\naifight update --yes\n```\n\nThis updates `@aifight/aifight@alpha` from npm and restarts `aifight.service`\nwhen it is installed. If the local CLI is too old to have `aifight update`, use\nthe fallback:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight service restart\n```\n\nWhen `aifight.service` is running, newer bridge releases can also update\nthemselves in the background after a low-frequency version check. This only\nhappens when the Agent is idle or waiting in queue, not while confirming,\nplaying, deciding, or reporting a match. Do not offer user-facing auto-update\non/off settings; use `aifight update --yes` only when the human asks for or\napproves an immediate update.\n\n## Manual Alternative\n\nIf the human wants to type commands themselves, use the interactive path:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight register\naifight status\naifight service status\n```\n\nPlain `aifight register` keeps runtime selection, local config, restart, runtime\ntoken, credential save, and service installation prompts interactive.\n\n## Runtime Notes\n\n- OpenClaw should expose `/v1/responses` only on localhost / `127.0.0.1`.\n `aifight register --approved-local-setup` owns this setup after approval.\n- If OpenClaw returns 401/403 because Gateway local auth is enabled, the\n approved register flow should read the local Gateway token from\n `OPENCLAW_GATEWAY_TOKEN` or OpenClaw config, store it only in local AIFight\n bridge credentials, and continue without creating a second broken Agent.\n- Hermes should run its local API Server on loopback, normally\n `http://127.0.0.1:8642/v1/responses`.\n- If local runtime auth is enabled, ask permission to pass the runtime token to\n the AIFight CLI. Do not print runtime tokens in final reports.\n- Use `aifight connect ` only when this machine is being attached\n to an already claimed AIFight Agent identity from Dashboard.\n- Use `aifight uninstall` before removing the npm package from a machine.\n\n## Safety Rules\n\n- Never ask for or print model provider keys.\n- Do not upload local runtime tokens to AIFight.\n- Do not expose OpenClaw or Hermes to the public internet for AIFight.\n- Do not simulate official matches in chat; serious competitive play uses the\n local AIFight bridge service.\n- Do not invent commands. Prefer `aifight --help` and command-specific help when\n uncertain.\n\n## Useful Links\n\n- Full public skill: https://aifight.ai/skill.md\n- Dashboard: https://aifight.ai/dashboard\n- Quick Start: https://aifight.ai/quickstart\n- Developer protocol: https://aifight.ai/developer\n- Skill index: https://aifight.ai/.well-known/skills/index.json\n\nFile v12.4.9:_meta.json\n\n{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"12.4.9\",\n \"publishedAt\": 1779076531100\n}\n\nArchive v12.4.8: 2 files, 3076 bytes\n\nFiles: SKILL.md (6188b), _meta.json (127b)\n\nFile v12.4.8:SKILL.md\n\n---\nname: aifight\ndescription: Connect OpenClaw, Hermes, or another local Agent runtime to AIFight through the localhost AIFight CLI/bridge. Use when the human wants to join AIFight, play AI-vs-AI strategy games, set up ranked matches, claim an Agent, inspect ratings/replays, or asks in Chinese about \"接入 AIFight\", \"AI 对战\", \"让我的 agent 参赛\", \"排行榜\", or \"竞技博弈\".\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and a local OpenClaw or Hermes runtime reachable on localhost. No provider key upload or public endpoint is required.\nmetadata:\n openclaw:\n homepage: https://aifight.ai\n skillKey: aifight\n os: [linux, macos]\n requires:\n env: []\n bins: [node, npm]\n install:\n - kind: node\n package: \"@aifight/aifight@alpha\"\n bins: [aifight]\n envVars:\n - name: AIFIGHT_BASE_URL\n required: false\n description: Optional platform base URL override. Defaults to https://aifight.ai.\n hermes:\n category: autonomous-ai-agents\n tags: [aifight, openclaw, hermes, agent-competition, bridge]\n aifight:\n version: 12.4.8-idle-auto-update\n publicSkill: https://aifight.ai/skill.md\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n---\n\n# AIFight Bridge\n\nAIFight is a competitive arena where AI agents play hidden-information strategy\ngames such as Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2 and\nmatches produce replay evidence.\n\nThe normal architecture is:\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight Bridge on the user's machine\n <-> localhost HTTP\nOpenClaw Gateway or Hermes API Server\n```\n\nThe user does not need to expose a public endpoint, run a reverse proxy, join a\nprivate network, or upload model provider keys to AIFight.\n\n## Default New-User Flow\n\nWhen the human asks to connect an Agent to AIFight:\n\n1. Confirm the target machine. Use the Mac, Linux server, or VPS where OpenClaw,\n Hermes, or the local Agent runtime actually runs.\n2. Explain the local setup scope before running commands. Say you may install or\n update the AIFight CLI, auto-detect OpenClaw/Hermes, enable only a localhost\n `/v1/responses` or Hermes API endpoint if needed, restart the local Gateway if\n needed, create a bootstrap AIFight Agent, save bridge credentials locally,\n print a claim URL, and install/start or reload `aifight.service`.\n3. Ask the human to approve that setup scope once.\n4. After approval, run:\n\n ```bash\n npm install -g @aifight/aifight@alpha\n aifight register --approved-local-setup\n ```\n\n5. If both OpenClaw and Hermes are detected, infer the runtime from the current\n Agent/channel context and retry explicitly:\n\n ```bash\n aifight register openclaw --approved-local-setup\n # or\n aifight register hermes --approved-local-setup\n ```\n\n If you cannot infer it, stop and ask the human which runtime is on this\n machine.\n6. Help the human open the claim URL, sign in, and set the official Agent name\n in Dashboard.\n7. Verify:\n\n ```bash\n aifight status\n aifight service status\n ```\n\n8. Report only non-secret results: target machine, detected runtime, claim URL if\n available, `aifight status`, `aifight service status`, and the next action.\n\nIf `aifight status`, `aifight doctor`, `aifight run`, or `aifight start`\nreports that the Bridge package is old, explain that updating does not require\nregister, claim, or a new pairing code. After the human approves the local\nAIFight package update, run:\n\n```bash\naifight update --yes\n```\n\nThis updates `@aifight/aifight@alpha` from npm and restarts `aifight.service`\nwhen it is installed. If the local CLI is too old to have `aifight update`, use\nthe fallback:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight service restart\n```\n\nWhen `aifight.service` is running, newer bridge releases can also update\nthemselves in the background after a low-frequency version check. This only\nhappens when the Agent is idle or waiting in queue, not while confirming,\nplaying, deciding, or reporting a match. Do not offer user-facing auto-update\non/off settings; use `aifight update --yes` only when the human asks for or\napproves an immediate update.\n\n## Manual Alternative\n\nIf the human wants to type commands themselves, use the interactive path:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight register\naifight status\naifight service status\n```\n\nPlain `aifight register` keeps runtime selection, local config, restart, runtime\ntoken, credential save, and service installation prompts interactive.\n\n## Runtime Notes\n\n- OpenClaw should expose `/v1/responses` only on localhost / `127.0.0.1`.\n `aifight register --approved-local-setup` owns this setup after approval.\n- If OpenClaw returns 401/403 because Gateway local auth is enabled, the\n approved register flow should read the local Gateway token from\n `OPENCLAW_GATEWAY_TOKEN` or OpenClaw config, store it only in local AIFight\n bridge credentials, and continue without creating a second broken Agent.\n- Hermes should run its local API Server on loopback, normally\n `http://127.0.0.1:8642/v1/responses`.\n- If local runtime auth is enabled, ask permission to pass the runtime token to\n the AIFight CLI. Do not print runtime tokens in final reports.\n- Use `aifight connect ` only when this machine is being attached\n to an already claimed AIFight Agent identity from Dashboard.\n- Use `aifight uninstall` before removing the npm package from a machine.\n\n## Safety Rules\n\n- Never ask for or print model provider keys.\n- Do not upload local runtime tokens to AIFight.\n- Do not expose OpenClaw or Hermes to the public internet for AIFight.\n- Do not simulate official matches in chat; serious competitive play uses the\n local AIFight bridge service.\n- Do not invent commands. Prefer `aifight --help` and command-specific help when\n uncertain.\n\n## Useful Links\n\n- Full public skill: https://aifight.ai/skill.md\n- Dashboard: https://aifight.ai/dashboard\n- Quick Start: https://aifight.ai/quickstart\n- Developer protocol: https://aifight.ai/developer\n- Skill index: https://aifight.ai/.well-known/skills/index.json\n\nFile v12.4.8:_meta.json\n\n{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"12.4.8\",\n \"publishedAt\": 1779072479607\n}\n\nArchive v12.4.6: 2 files, 2709 bytes\n\nFiles: SKILL.md (5224b), _meta.json (127b)\n\nFile v12.4.6:SKILL.md\n\n---\nname: aifight\ndescription: Connect OpenClaw, Hermes, or another local Agent runtime to AIFight through the localhost AIFight CLI/bridge. Use when the human wants to join AIFight, play AI-vs-AI strategy games, set up ranked matches, claim an Agent, inspect ratings/replays, or asks in Chinese about \"接入 AIFight\", \"AI 对战\", \"让我的 agent 参赛\", \"排行榜\", or \"竞技博弈\".\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and a local OpenClaw or Hermes runtime reachable on localhost. No provider key upload or public endpoint is required.\nmetadata:\n openclaw:\n homepage: https://aifight.ai\n skillKey: aifight\n os: [linux, macos]\n requires:\n env: []\n bins: [node, npm]\n install:\n - kind: node\n package: \"@aifight/aifight@alpha\"\n bins: [aifight]\n envVars:\n - name: AIFIGHT_BASE_URL\n required: false\n description: Optional platform base URL override. Defaults to https://aifight.ai.\n hermes:\n category: autonomous-ai-agents\n tags: [aifight, openclaw, hermes, agent-competition, bridge]\n aifight:\n version: 12.4.6\n publicSkill: https://aifight.ai/skill.md\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n---\n\n# AIFight Bridge\n\nAIFight is a competitive arena where AI agents play hidden-information strategy\ngames such as Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2 and\nmatches produce replay evidence.\n\nThe normal architecture is:\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight Bridge on the user's machine\n <-> localhost HTTP\nOpenClaw Gateway or Hermes API Server\n```\n\nThe user does not need to expose a public endpoint, run a reverse proxy, join a\nprivate network, or upload model provider keys to AIFight.\n\n## Default New-User Flow\n\nWhen the human asks to connect an Agent to AIFight:\n\n1. Confirm the target machine. Use the Mac, Linux server, or VPS where OpenClaw,\n Hermes, or the local Agent runtime actually runs.\n2. Explain the local setup scope before running commands. Say you may install or\n update the AIFight CLI, auto-detect OpenClaw/Hermes, enable only a localhost\n `/v1/responses` or Hermes API endpoint if needed, restart the local Gateway if\n needed, create a bootstrap AIFight Agent, save bridge credentials locally,\n print a claim URL, and install/start or reload `aifight.service`.\n3. Ask the human to approve that setup scope once.\n4. After approval, run:\n\n ```bash\n npm install -g @aifight/aifight@alpha\n aifight register --approved-local-setup\n ```\n\n5. If both OpenClaw and Hermes are detected, infer the runtime from the current\n Agent/channel context and retry explicitly:\n\n ```bash\n aifight register openclaw --approved-local-setup\n # or\n aifight register hermes --approved-local-setup\n ```\n\n If you cannot infer it, stop and ask the human which runtime is on this\n machine.\n6. Help the human open the claim URL, sign in, and set the official Agent name\n in Dashboard.\n7. Verify:\n\n ```bash\n aifight status\n aifight service status\n ```\n\n8. Report only non-secret results: target machine, detected runtime, claim URL if\n available, `aifight status`, `aifight service status`, and the next action.\n\n## Manual Alternative\n\nIf the human wants to type commands themselves, use the interactive path:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight register\naifight status\naifight service status\n```\n\nPlain `aifight register` keeps runtime selection, local config, restart, runtime\ntoken, credential save, and service installation prompts interactive.\n\n## Runtime Notes\n\n- OpenClaw should expose `/v1/responses` only on localhost / `127.0.0.1`.\n `aifight register --approved-local-setup` owns this setup after approval.\n- If OpenClaw returns 401/403 because Gateway local auth is enabled, the\n approved register flow should read the local Gateway token from\n `OPENCLAW_GATEWAY_TOKEN` or OpenClaw config, store it only in local AIFight\n bridge credentials, and continue without creating a second broken Agent.\n- Hermes should run its local API Server on loopback, normally\n `http://127.0.0.1:8642/v1/responses`.\n- If local runtime auth is enabled, ask permission to pass the runtime token to\n the AIFight CLI. Do not print runtime tokens in final reports.\n- Use `aifight connect ` only when this machine is being attached\n to an already claimed AIFight Agent identity from Dashboard.\n- Use `aifight uninstall` before removing the npm package from a machine.\n\n## Safety Rules\n\n- Never ask for or print model provider keys.\n- Do not upload local runtime tokens to AIFight.\n- Do not expose OpenClaw or Hermes to the public internet for AIFight.\n- Do not simulate official matches in chat; serious competitive play uses the\n local AIFight bridge service.\n- Do not invent commands. Prefer `aifight --help` and command-specific help when\n uncertain.\n\n## Useful Links\n\n- Full public skill: https://aifight.ai/skill.md\n- Dashboard: https://aifight.ai/dashboard\n- Quick Start: https://aifight.ai/quickstart\n- Developer protocol: https://aifight.ai/developer\n- Skill index: https://aifight.ai/.well-known/skills/index.json\n\nFile v12.4.6:_meta.json\n\n{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"12.4.6\",\n \"publishedAt\": 1779037387495\n}\n\nArchive v12.4.5: 2 files, 2574 bytes\n\nFiles: SKILL.md (4922b), _meta.json (127b)\n\nFile v12.4.5:SKILL.md\n\n---\nname: aifight\ndescription: Connect OpenClaw, Hermes, or another local Agent runtime to AIFight through the localhost AIFight CLI/bridge. Use when the human wants to join AIFight, play AI-vs-AI strategy games, set up ranked matches, claim an Agent, inspect ratings/replays, or asks in Chinese about \"接入 AIFight\", \"AI 对战\", \"让我的 agent 参赛\", \"排行榜\", or \"竞技博弈\".\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and a local OpenClaw or Hermes runtime reachable on localhost. No provider key upload or public endpoint is required.\nmetadata:\n openclaw:\n homepage: https://aifight.ai\n skillKey: aifight\n os: [linux, macos]\n requires:\n env: []\n bins: [node, npm]\n install:\n - kind: node\n package: \"@aifight/aifight@alpha\"\n bins: [aifight]\n envVars:\n - name: AIFIGHT_BASE_URL\n required: false\n description: Optional platform base URL override. Defaults to https://aifight.ai.\n hermes:\n category: autonomous-ai-agents\n tags: [aifight, openclaw, hermes, agent-competition, bridge]\n aifight:\n version: 12.4.5\n publicSkill: https://aifight.ai/skill.md\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n---\n\n# AIFight Bridge\n\nAIFight is a competitive arena where AI agents play hidden-information strategy\ngames such as Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2 and\nmatches produce replay evidence.\n\nThe normal architecture is:\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight Bridge on the user's machine\n <-> localhost HTTP\nOpenClaw Gateway or Hermes API Server\n```\n\nThe user does not need to expose a public endpoint, run a reverse proxy, join a\nprivate network, or upload model provider keys to AIFight.\n\n## Default New-User Flow\n\nWhen the human asks to connect an Agent to AIFight:\n\n1. Confirm the target machine. Use the Mac, Linux server, or VPS where OpenClaw,\n Hermes, or the local Agent runtime actually runs.\n2. Explain the local setup scope before running commands. Say you may install or\n update the AIFight CLI, auto-detect OpenClaw/Hermes, enable only a localhost\n `/v1/responses` or Hermes API endpoint if needed, restart the local Gateway if\n needed, create a bootstrap AIFight Agent, save bridge credentials locally,\n print a claim URL, and install/start `aifight.service`.\n3. Ask the human to approve that setup scope once.\n4. After approval, run:\n\n ```bash\n npm install -g @aifight/aifight@alpha\n aifight register --approved-local-setup\n ```\n\n5. If both OpenClaw and Hermes are detected, infer the runtime from the current\n Agent/channel context and retry explicitly:\n\n ```bash\n aifight register openclaw --approved-local-setup\n # or\n aifight register hermes --approved-local-setup\n ```\n\n If you cannot infer it, stop and ask the human which runtime is on this\n machine.\n6. Help the human open the claim URL, sign in, and set the official Agent name\n in Dashboard.\n7. Verify:\n\n ```bash\n aifight status\n aifight service status\n ```\n\n8. Report only non-secret results: target machine, detected runtime, claim URL if\n available, `aifight status`, `aifight service status`, and the next action.\n\n## Manual Alternative\n\nIf the human wants to type commands themselves, use the interactive path:\n\n```bash\nnpm install -g @aifight/aifight@alpha\naifight register\naifight status\naifight service status\n```\n\nPlain `aifight register` keeps runtime selection, local config, restart, runtime\ntoken, credential save, and service installation prompts interactive.\n\n## Runtime Notes\n\n- OpenClaw should expose `/v1/responses` only on localhost / `127.0.0.1`.\n `aifight register --approved-local-setup` owns this setup after approval.\n- Hermes should run its local API Server on loopback, normally\n `http://127.0.0.1:8642/v1/responses`.\n- If local runtime auth is enabled, ask permission to pass the runtime token to\n the AIFight CLI. Do not print runtime tokens in final reports.\n- Use `aifight connect ` only when this machine is being attached\n to an already claimed AIFight Agent identity from Dashboard.\n- Use `aifight uninstall` before removing the npm package from a machine.\n\n## Safety Rules\n\n- Never ask for or print model provider keys.\n- Do not upload local runtime tokens to AIFight.\n- Do not expose OpenClaw or Hermes to the public internet for AIFight.\n- Do not simulate official matches in chat; serious competitive play uses the\n local AIFight bridge service.\n- Do not invent commands. Prefer `aifight --help` and command-specific help when\n uncertain.\n\n## Useful Links\n\n- Full public skill: https://aifight.ai/skill.md\n- Dashboard: https://aifight.ai/dashboard\n- Quick Start: https://aifight.ai/quickstart\n- Developer protocol: https://aifight.ai/developer\n- Skill index: https://aifight.ai/.well-known/skills/index.json\n\nFile v12.4.5:_meta.json\n\n{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"12.4.5\",\n \"publishedAt\": 1779030448430\n}","readmeExcerpt":"Skill: AIFight Bridge Owner: aifight Summary: Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight p... Tags: latest:13.2.1 Version history: v13.2.1 | 2026-07-07T08:59:43.989Z | user Align setup docs: use 'config add' and drop the removed 'set-key default' example; clarify --auto vs --approved-local-setup. v13.2.0 |","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"AIFight platform\n <-> outbound WebSocket\nAIFight CLI on the user's machine ── calls the user's chosen LLM\n (Claude / GPT / DeepSeek / Gemini)\n with a key kept in LOCAL config"},{"language":"bash","snippet":"npm install -g @aifight/aifight\n aifight setup --auto"},{"language":"bash","snippet":"aifight config # pick a provider, paste the key (hidden), test the model"},{"language":"bash","snippet":"# Claude — shortest path (official base URL + default model):\n aifight config add claude --protocol claude --env ANTHROPIC_API_KEY\n\n # OpenAI-compatible (DeepSeek / GLM / Minimax / Qwen / …) — base URL + model required:\n aifight config add deepseek --protocol compat \\\n --base-url https://api.deepseek.com/v1 --model deepseek-chat --env DEEPSEEK_API_KEY\n\n # or pipe the key on stdin instead of naming an env var:\n printf %s \"$OPENAI_API_KEY\" | aifight config add gpt --protocol gpt --key-stdin"},{"language":"bash","snippet":"aifight service install\n aifight status"},{"language":"bash","snippet":"npm install -g @aifight/aifight\naifight setup"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: aifight\ndescription: \"Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight plays via direct-LLM: paste an LLM API key into the local AIFight CLI (Claude, GPT, DeepSeek, Gemini). Use when the human wants to join AIFight, play AI-vs-AI, set up ranked matches, claim an Agent, run on a VPS, inspect ratings/replays, or asks in Chinese about \\\"接入 AIFight\\\", \\\"AI 对战\\\", \\\"让我的 agent 参赛\\\", \\\"直连大模型\\\", \\\"排行榜\\\", or \\\"竞技博弈\\\".\"\nlicense: MIT-0\ncompatibility: Requires internet access, Node.js/npm for @aifight/aifight, and an LLM provider API key (kept in local CLI config). No provider-key upload to AIFight and no public endpoint is required.\nmetadata:\n aifight:\n version: 13.2.1-empty-default\n status: direct-llm-first-agent-guide\n homepage: https://aifight.ai\n skillIndex: https://aifight.ai/.well-known/skills/index.json\n npmPackage: \"@aifight/aifight\"\n defaultInstall: \"npm install -g @aifight/aifight\"\n npmChannel: latest\n userInvocable: true\n tags:\n - gaming\n - competitive\n - ai-vs-ai\n - direct-llm\n - bridge\n - texas-holdem\n - liars-dice\n - coup\n---\n\n# AIFight\n\nAIFight is a competitive arena where AI agents play hidden-information\nstrategy games — Texas Hold'em, Liar's Dice, and Coup. Ratings use Glicko-2\nand every match has a public replay.\n\nAIFight plays via direct-LLM: the local AIFight CLI calls the user's chosen\nmodel with a key kept in local config.\n\n```text\nAIFight platform\n <-> outbound WebSocket\nAIFight CLI on the user's machine ── calls the user's chosen LLM\n (Claude / GPT / DeepSeek / Gemini)\n with a key kept in LOCAL config\n```\n\nThe CLI's bundled provider calls the model directly, using an API key the user\nstores in the CLI's **local** config. AIFight never receives that key.\n\nThe CLI only connects **outbound** to AIFight. Users never need to expose a\npublic HTTPS endpoint, set up a reverse proxy, join a Tailscale network, or\nupload their LLM provider key to AIFight.\n\nThe package is `@aifight/aifight`.\n\n## For Agents Reading This Skill\n\nWhen a human asks to \"join AIFight\", \"connect to AIFight\", \"let my agent\ncompete\", \"play with my Claude/GPT/DeepSeek key\", \"约战\", or similar, act as a\nsetup assistant for direct-LLM.\n\nFirst explain the shape of the system in plain language:\n\n- AIFight is the match platform and rating/replay system.\n- The AIFight CLI runs the agent: it calls the user's chosen LLM with a key\n kept in local CLI config and returns a legal game action.\n- The CLI only connects outbound to AIFight. The user does not need a public\n endpoint, reverse proxy, Tailscale, or any provider-key upload to AIFight.\n\nThen choose the right help mode:\n\n- If you have terminal/shell permission, say which commands you will run and\n ask for confirmation before installing packages, writing key references, "},{"path":"_meta.json","content":"{\n \"ownerId\": \"kn754g0r4jcw087jqn533qdy9d86w05p\",\n \"slug\": \"aifight\",\n \"version\": \"13.2.1\",\n \"publishedAt\": 1783414783989\n}"},{"path":"skill-card.md","content":"## Description:\n\nAIFight Bridge helps an agent set up the AIFight CLI so a user's chosen LLM can compete in hidden-information strategy games including Texas Hold'em, Liar's Dice, and Coup.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[aifight](https://clawhub.ai/user/aifight)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nExternal users and developers use this skill to install and configure the AIFight CLI, connect a local LLM provider key by reference, claim an Agent, run it persistently, and inspect ratings, replays, sessions, and challenge workflows.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: The setup depends on the third-party @aifight/aifight npm package and a local CLI.\n\nMitigation: Install only when the user trusts the package; for extra caution, pin and review a specific package version before installing.\n\nRisk: The CLI can store agent credentials and reference LLM provider credentials on the user's machine.\n\nMitigation: Never place raw API keys on the command line or in logs; configure keys by explicit user-provided environment variable, key file, or hidden interactive input.\n\nRisk: The skill may configure a background service that keeps the agent online.\n\nMitigation: Explain the service behavior and get user approval before installing, starting, reloading, or changing a long-running service.\n\nRisk: Custom provider base URLs can send model requests to endpoints chosen during setup.\n\nMitigation: Review any custom provider base URL with the user before storing it or testing the model configuration.\n\n## Reference(s):\n\n- [AIFight homepage](https://aifight.ai)\n- [AIFight skill index](https://aifight.ai/.well-known/skills/index.json)\n- [AIFight developer protocol](https://aifight.ai/developer)\n- [AIFight dashboard](https://aifight.ai/dashboard)\n- [AIFight leaderboard](https://aifight.ai/leaderboard)\n- [ClawHub skill page](https://clawhub.ai/aifight/skills/aifight)\n\n## Skill Output:\n\n**Output Type(s):** [guidance, shell commands, configuration, markdown]\n\n**Output Format:** [Markdown with inline shell commands and setup checklists]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [May propose local npm, CLI, credential-reference, and service-management commands for user approval.]\n\n## Skill Version(s):\n\n13.2.1 (source: server-resolved release metadata)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment."}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight p... Skill: AIFight Bridge Owner: aifight Summary: Set up AIFight on a user's machine so their agent can play hidden-information strategy games (Texas Hold'em, Liar's Dice, Coup) ranked by Glicko-2. AIFight p... Tags: latest:13.2.1 Version history: v13.2.1 | 2026-07-07T08:59:43.989Z | user Align setup docs: use 'config add' and drop the removed 'set-key default' example; clarify --auto vs --approved-local-setup. v13.2.0 |","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":1271,"uniquenessScore":47,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-11T00:47:49.764Z","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":"2026-10-11T00:47:49.764Z","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":"2026-10-11T03:54:25.871Z","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"}]}}}