[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6HYFSG3M2Q8FS86K7AEF0HN":3,"skill-same:01M6HYFSG3M2Q8FS86K7AEF0HN":60},{"skillNo":4,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":21,"linkOnly":22,"markdownContent":23,"markdownBody":24,"frontMatter":25,"refName":26,"commitSha":27,"skillDirectory":28,"forksCount":13,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M6HYFSG3M2Q8FS86K7AEF0HN","discover-twitter-recruits",null,"Jinn-Network\u002Fmono","Jinn-Network","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F236846593?v=4","01M6HYDXMQQYRMV3QZPT7JYY87","Jinn Network","01M6HYDXS833N3PY8EJ5NHJGB6",2,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Fgrowth\u002Farchive\u002Fskills\u002Fdiscover-twitter-recruits\u002FSKILL.md",[],[18,19,20],"twitter","discover","discover twitter recruits","2026-11-10 10:12:30",false,"---\nname: discover-twitter-recruits\ndescription: Use when the user wants to find Twitter accounts to recruit into Jinn — operators or builders adjacent to OLAS, Pearl, Bittensor, Numerai, ERC-8004, agent verification, agent observability, agent evaluation, or prediction-tooling work. Triggers on \"find people to talk to about Jinn\", \"who's posting about agent verification \u002F olas \u002F pearl \u002F forecasting \u002F bittensor \u002F numerai \u002F erc-8004\", \"discover Twitter recruits\", \"find more accounts like @yieldfreaks\", \"find more like silverarrow\", or any variation that implies surfacing candidate accounts rather than engaging them. Also triggers when the user is at risk of drafting outreach to an account whose audience-fit has not been profile-checked. Out of scope: drafting reply text, planning engagement sequences, recommending follow-ups.\n---\n\n# Discover Twitter recruits\n\nFind Twitter accounts that match the audience Jinn is recruiting. Output is a list of candidates with rationale and a reply-rate ranking. Engagement and outreach are explicitly out of scope.\n\n## What this skill does\n\nOne mode: **discover**. The user names a topic, an audience description, or asks for \"more like X\"; the skill produces a triaged list. The skill does not draft outreach. If the user asks for that, defer — say so plainly and ask whether they want a separate engagement skill or this skill's output as input to one.\n\n## The mental model in one paragraph\n\nDiscovery is an audience filter, not a language filter. The right question for every candidate is: *would this person plausibly run a Jinn solver, build adjacent tooling, contribute code, or boost the protocol with intellectual reach?* Thesis-shaped tweets from enterprise consultants, JS framework engineers, agent-memory tool vendors, or token-pumping crypto-AI shillers do not pass — even when their wording matches THESIS.md word-for-word. The signal is whether the account *ships verifiable artefacts in the right orbit*, not whether it talks the right way.\n\nFor the conversion criterion in detail, read `references\u002Faudience-profile.md`. For working and failed search vocabularies, read `references\u002Fsearch-strategy.md`. After every invocation that leads to outreach, append to `references\u002Fdiscovery-log.md`.\n\n## When you run\n\nRead `references\u002Faudience-profile.md` and `references\u002Fsearch-strategy.md` once at the start of a session. The audience profile defines the conversion bar; the search strategy defines the working `bird` CLI invocations and the anti-patterns to avoid. Both are short.\n\nThe skill assumes the `bird` CLI is available. If it is not, ask the user to install it and stop — this skill does not fall back to web search, because the failure mode is too easy (search engines surface high-reach tweets, which over-represents shills and large accounts).\n\n## Discovery procedure\n\nApply in order. Stop early if the search consistently returns the wrong shape.\n\n### Step 1 — Frame the audience\n\nBefore any search, write down (internally) the specific audience for this invocation. Defaults derive from [`GROWTH.md`](..\u002F..\u002F..\u002FGROWTH.md) §3 (current target cluster) and §4 (GTM phase clusters, sampled lighter when active sprint allows):\n\n- **Primary** — current target cluster from GROWTH §3.\n- **Adjacent (Phase 2 \u002F Phase 3)** — GROWTH §4 phases sampled only when the active sprint or refine-skill output flags a phase-transition check.\n- **Amplifiers (separate tier)** — builders with primitives-not-platforms instincts who repost rather than operate. Output as a separate bucket; conversion shape differs (a quote-tweet, not a node).\n\nIf the user names a topic that does not map to GROWTH §3 \u002F §4, ask once which audience they are aiming at before searching. Wrong-audience candidates burn discovery budget.\n\n### Step 2 — Search with builder vocabulary\n\nRun a small batch of `bird search` queries using *builder-shaped* terms (audience-profile §1, search-strategy §2):\n\n- Builder vocabulary: `agent registry`, `agent observability`, `agent benchmarking`, `evaluation framework`, `eval signal`, `ground truth`.\n- Audience-name vocabulary: `olas pearl`, `olas mech operator`, `bittensor subnet operator`, `polymarket bot`, `prediction subnet`, `numinous`, `autoharness`.\n- Cross-reference: replies to `@autonolas`, mentions of `@numinous_ai`, `@opentensor`, replies under specific a16z crypto threads.\n\nAvoid the named anti-pattern queries (search-strategy §1) — they catch shills and bots, not builders.\n\n### Step 3 — Profile-check each candidate\n\nBefore recommending any account, run all three:\n\n- `bird user-tweets \u003Chandle>` — read the most recent ~5–10 tweets. Confirm a *posting pattern*, not a single tweet that happened to match. A real builder posts about their work over weeks; a shill posts identical-shape promotional content; a bot posts one-shot zingers with no thread engagement.\n- `bird about \u003Chandle>` — sanity-check geographic and creation signals. Not a hard rule, but high signal when paired with red flags.\n- Confirm a real artefact in the bio or recent posts: linked repo, dashboard, dataset, product, or paper. Pure takes-only accounts go to the *amplifiers* tier, not the main list.\n\n### Step 4 — Bot\u002Fshill detection\n\nReject (do not recommend, do not amplify) any account matching these patterns:\n\n- `?` sign-off — strong OpenClaw agent signal. Confirm by checking thread engagement: bots reply to no one.\n- One-shot zinger pattern — every post is a contrarian-take fragment with no follow-up replies, no QTs of others, no conversation.\n- Hashtag spam — `#AgenticAI #AI #ML #Web3` stacks indicate low-effort or automated posting.\n- Marketing register — \"we are so early\", \"this is a game changer\", \"the future is...\", \"bullish doesn't even cover it\".\n- Token-ticker preludes — posts that start with `$XYZ` or end with `CA: 0x...`.\n- Identical-shape posts across many accounts — symptom of shill ring.\n\nA candidate that fails any of these does not appear in the output, even if a single tweet was thesis-perfect. (Lesson logged in `references\u002Fdiscovery-log.md` from the `@gingersamurai` correction.)\n\n### Step 5 — Output\n\nUse the exact structure below. Order candidates by reply-rate probability — most likely to engage first.\n\n## Output format\n\n```\nDISCOVERY: \u003Cone-line description of who was being looked for>\n\nOPERATORS \u002F BUILDERS \u002F CONTRIBUTORS\n\nN. @handle (Display Name)\n- URL: \u003Clink to most recent on-thesis post>\n- Who they are: \u003Cone sentence — actual product, role, or output>\n- Recent on-thesis post: \"\u003Cshort verbatim quote>\"\n- Why they convert: \u003Cone sentence — operator \u002F builder \u002F contributor and why>\n- Thesis link: \u003Cwhich section of THESIS.md or GROWTH.md they map to>\n- Reply-rate signal: \u003Cengagement count on recent post \u002F account size class \u002F time-since-post>\n\n[repeat for each candidate]\n\nAMPLIFIERS (separate tier — boosters with reach, not operators)\n\nN. @handle (Display Name)\n- URL, Who they are, Recent post, Why they amplify, Reply-rate signal — same fields, different framing.\n\nREPLY-RATE RANKING\n1. @handle — \u003Cone-line reason this is the warmest first contact>\n2. @handle — \u003Creason>\n...\n\nSKIPPED (audit trail — candidates considered and rejected)\n- @handle — \u003Cone-line reason: wrong audience \u002F no real artefact \u002F bot pattern \u002F etc.>\n- @handle — \u003Creason>\n...\n```\n\nHonesty rules:\n\n- If only 3 candidates pass the bar, return 3 and say so. Do not pad.\n- The SKIPPED section is mandatory — it preserves the audit trail and prevents the same wrong shapes from being re-recommended.\n- The reply-rate ranking is a probability estimate, not a recommendation to outreach. The skill produces the list; the user (or a separate skill) decides what to do with it.\n\n## Voice constraints\n\n- British English.\n- No emoji.\n- Builder-to-builder vocabulary. \"Ships X\" not \"leverages X\". \"Built Y\" not \"unlocks Y\". Strip marketing register completely; if a sentence reads as if it could appear on a project's landing page, rewrite it.\n- Plain prose; lists only where structure earns its keep.\n- \"Honest 3\" beats \"padded 5\" — do not invent candidates to hit a target count.\n\n## Calibration loop\n\nAfter any invocation that leads to outreach, append to `references\u002Fdiscovery-log.md`:\n\n- Candidate handle.\n- Search query that surfaced them.\n- Conversion rationale at time of recommendation.\n- Outreach path (which post replied to, which question asked).\n- Outcome: did they reply, what did they say, did they engage further.\n- Lesson, if any: a tweak to `audience-profile.md` or `search-strategy.md`.\n\nAfter 5–10 entries, review whether the audience profile or search vocabulary needs adjustment. The model is a starting point; the discovery log is what makes it Jinn's recruitment lattice rather than a generic Twitter-search guide.\n\n## Two logs — calibration vs operational\n\nThis skill maintains two logs with distinct purposes:\n\n- **`references\u002Fdiscovery-log.md`** — skill calibration evidence. Recommendation rationale + outcome + lesson, kept in repo so the skill learns over time. Anonymise where the lesson is generalisable. Update this file after any outreach attempt that produces a clear lesson.\n- **`growth\u002F.local\u002Fgrowth-log.md`** — operational state. Live thread state, drafts, pending replies, candidate handles in operational form. Gitignored. Updated by Oak and by `growth-watcher` \u002F `growth-day` skills.\n\nDo not write live operational state into `discovery-log.md`. Do not duplicate calibration lessons into `growth-log.md`. The boundary is sharp because the failure mode of conflating them is real — calibration evidence in the operational log produces noise; operational state in the calibration log produces leakage.\n\n## Out of scope\n\nThis skill does not:\n\n- Draft reply text or DM messages.\n- Plan engagement sequences (first \u002F second \u002F third touch).\n- Suggest off-platform contact paths.\n- Recommend follow-ups after silence.\n- Score posts for posting (that is `x-algorithm-grader`).\n- Generate posts (that is a future content skill).\n\nIf the user wants any of the above, return the list and surface the boundary plainly. The discovery output is the input to those skills, not their replacement.\n\n## Composing with other skills\n\n- `x-algorithm-grader` — for grading the user's *own* drafts (not for grading candidates' posts; cluster fit is a different question for posts vs people).\n- A future `x-engagement-planner` skill (not yet built) would consume the output of this skill and produce per-candidate first-touch sequences, second-touch contingencies, and reply-rate-aware cadence.\n\n## Reference files\n\n- `references\u002Faudience-profile.md` — the conversion criterion; who counts as a recruit, who does not, with one-line reasons for each excluded audience.\n- `references\u002Fsearch-strategy.md` — concrete `bird` CLI invocations, working vocabularies, and named anti-patterns.\n- `references\u002Fdiscovery-log.md` — empirical record of past discoveries, started from the @yieldfreaks \u002F @ta_eis_eauton \u002F @gingersamurai cases.\n\n## Why this skill exists\n\nTwo recent successful recruitments (`@yieldfreaks`, builder of an agent-registry health dashboard; `@ta_eis_eauton` \u002F Silverarrow, builder of an open-source agent-evaluation harness) shared a shape that was not visible until both had been found: small\u002Fmedium accounts shipping verifiable tooling adjacent to verification, evaluation, or registry infra in the OLAS \u002F Bittensor \u002F Numerai orbits. Without this skill the next discovery attempt would default to thesis-language matching, which earlier in the same session produced a list of enterprise CIOs, JS framework devs, and an OpenClaw agent. Capturing the working filter as a skill makes the next round repeatable and prevents the regression.\n","# Discover Twitter recruits\n\nFind Twitter accounts that match the audience Jinn is recruiting. Output is a list of candidates with rationale and a reply-rate ranking. Engagement and outreach are explicitly out of scope.\n\n## What this skill does\n\nOne mode: **discover**. The user names a topic, an audience description, or asks for \"more like X\"; the skill produces a triaged list. The skill does not draft outreach. If the user asks for that, defer — say so plainly and ask whether they want a separate engagement skill or this skill's output as input to one.\n\n## The mental model in one paragraph\n\nDiscovery is an audience filter, not a language filter. The right question for every candidate is: *would this person plausibly run a Jinn solver, build adjacent tooling, contribute code, or boost the protocol with intellectual reach?* Thesis-shaped tweets from enterprise consultants, JS framework engineers, agent-memory tool vendors, or token-pumping crypto-AI shillers do not pass — even when their wording matches THESIS.md word-for-word. The signal is whether the account *ships verifiable artefacts in the right orbit*, not whether it talks the right way.\n\nFor the conversion criterion in detail, read `references\u002Faudience-profile.md`. For working and failed search vocabularies, read `references\u002Fsearch-strategy.md`. After every invocation that leads to outreach, append to `references\u002Fdiscovery-log.md`.\n\n## When you run\n\nRead `references\u002Faudience-profile.md` and `references\u002Fsearch-strategy.md` once at the start of a session. The audience profile defines the conversion bar; the search strategy defines the working `bird` CLI invocations and the anti-patterns to avoid. Both are short.\n\nThe skill assumes the `bird` CLI is available. If it is not, ask the user to install it and stop — this skill does not fall back to web search, because the failure mode is too easy (search engines surface high-reach tweets, which over-represents shills and large accounts).\n\n## Discovery procedure\n\nApply in order. Stop early if the search consistently returns the wrong shape.\n\n### Step 1 — Frame the audience\n\nBefore any search, write down (internally) the specific audience for this invocation. Defaults derive from [`GROWTH.md`](..\u002F..\u002F..\u002FGROWTH.md) §3 (current target cluster) and §4 (GTM phase clusters, sampled lighter when active sprint allows):\n\n- **Primary** — current target cluster from GROWTH §3.\n- **Adjacent (Phase 2 \u002F Phase 3)** — GROWTH §4 phases sampled only when the active sprint or refine-skill output flags a phase-transition check.\n- **Amplifiers (separate tier)** — builders with primitives-not-platforms instincts who repost rather than operate. Output as a separate bucket; conversion shape differs (a quote-tweet, not a node).\n\nIf the user names a topic that does not map to GROWTH §3 \u002F §4, ask once which audience they are aiming at before searching. Wrong-audience candidates burn discovery budget.\n\n### Step 2 — Search with builder vocabulary\n\nRun a small batch of `bird search` queries using *builder-shaped* terms (audience-profile §1, search-strategy §2):\n\n- Builder vocabulary: `agent registry`, `agent observability`, `agent benchmarking`, `evaluation framework`, `eval signal`, `ground truth`.\n- Audience-name vocabulary: `olas pearl`, `olas mech operator`, `bittensor subnet operator`, `polymarket bot`, `prediction subnet`, `numinous`, `autoharness`.\n- Cross-reference: replies to `@autonolas`, mentions of `@numinous_ai`, `@opentensor`, replies under specific a16z crypto threads.\n\nAvoid the named anti-pattern queries (search-strategy §1) — they catch shills and bots, not builders.\n\n### Step 3 — Profile-check each candidate\n\nBefore recommending any account, run all three:\n\n- `bird user-tweets \u003Chandle>` — read the most recent ~5–10 tweets. Confirm a *posting pattern*, not a single tweet that happened to match. A real builder posts about their work over weeks; a shill posts identical-shape promotional content; a bot posts one-shot zingers with no thread engagement.\n- `bird about \u003Chandle>` — sanity-check geographic and creation signals. Not a hard rule, but high signal when paired with red flags.\n- Confirm a real artefact in the bio or recent posts: linked repo, dashboard, dataset, product, or paper. Pure takes-only accounts go to the *amplifiers* tier, not the main list.\n\n### Step 4 — Bot\u002Fshill detection\n\nReject (do not recommend, do not amplify) any account matching these patterns:\n\n- `?` sign-off — strong OpenClaw agent signal. Confirm by checking thread engagement: bots reply to no one.\n- One-shot zinger pattern — every post is a contrarian-take fragment with no follow-up replies, no QTs of others, no conversation.\n- Hashtag spam — `#AgenticAI #AI #ML #Web3` stacks indicate low-effort or automated posting.\n- Marketing register — \"we are so early\", \"this is a game changer\", \"the future is...\", \"bullish doesn't even cover it\".\n- Token-ticker preludes — posts that start with `$XYZ` or end with `CA: 0x...`.\n- Identical-shape posts across many accounts — symptom of shill ring.\n\nA candidate that fails any of these does not appear in the output, even if a single tweet was thesis-perfect. (Lesson logged in `references\u002Fdiscovery-log.md` from the `@gingersamurai` correction.)\n\n### Step 5 — Output\n\nUse the exact structure below. Order candidates by reply-rate probability — most likely to engage first.\n\n## Output format\n\n```\nDISCOVERY: \u003Cone-line description of who was being looked for>\n\nOPERATORS \u002F BUILDERS \u002F CONTRIBUTORS\n\nN. @handle (Display Name)\n- URL: \u003Clink to most recent on-thesis post>\n- Who they are: \u003Cone sentence — actual product, role, or output>\n- Recent on-thesis post: \"\u003Cshort verbatim quote>\"\n- Why they convert: \u003Cone sentence — operator \u002F builder \u002F contributor and why>\n- Thesis link: \u003Cwhich section of THESIS.md or GROWTH.md they map to>\n- Reply-rate signal: \u003Cengagement count on recent post \u002F account size class \u002F time-since-post>\n\n[repeat for each candidate]\n\nAMPLIFIERS (separate tier — boosters with reach, not operators)\n\nN. @handle (Display Name)\n- URL, Who they are, Recent post, Why they amplify, Reply-rate signal — same fields, different framing.\n\nREPLY-RATE RANKING\n1. @handle — \u003Cone-line reason this is the warmest first contact>\n2. @handle — \u003Creason>\n...\n\nSKIPPED (audit trail — candidates considered and rejected)\n- @handle — \u003Cone-line reason: wrong audience \u002F no real artefact \u002F bot pattern \u002F etc.>\n- @handle — \u003Creason>\n...\n```\n\nHonesty rules:\n\n- If only 3 candidates pass the bar, return 3 and say so. Do not pad.\n- The SKIPPED section is mandatory — it preserves the audit trail and prevents the same wrong shapes from being re-recommended.\n- The reply-rate ranking is a probability estimate, not a recommendation to outreach. The skill produces the list; the user (or a separate skill) decides what to do with it.\n\n## Voice constraints\n\n- British English.\n- No emoji.\n- Builder-to-builder vocabulary. \"Ships X\" not \"leverages X\". \"Built Y\" not \"unlocks Y\". Strip marketing register completely; if a sentence reads as if it could appear on a project's landing page, rewrite it.\n- Plain prose; lists only where structure earns its keep.\n- \"Honest 3\" beats \"padded 5\" — do not invent candidates to hit a target count.\n\n## Calibration loop\n\nAfter any invocation that leads to outreach, append to `references\u002Fdiscovery-log.md`:\n\n- Candidate handle.\n- Search query that surfaced them.\n- Conversion rationale at time of recommendation.\n- Outreach path (which post replied to, which question asked).\n- Outcome: did they reply, what did they say, did they engage further.\n- Lesson, if any: a tweak to `audience-profile.md` or `search-strategy.md`.\n\nAfter 5–10 entries, review whether the audience profile or search vocabulary needs adjustment. The model is a starting point; the discovery log is what makes it Jinn's recruitment lattice rather than a generic Twitter-search guide.\n\n## Two logs — calibration vs operational\n\nThis skill maintains two logs with distinct purposes:\n\n- **`references\u002Fdiscovery-log.md`** — skill calibration evidence. Recommendation rationale + outcome + lesson, kept in repo so the skill learns over time. Anonymise where the lesson is generalisable. Update this file after any outreach attempt that produces a clear lesson.\n- **`growth\u002F.local\u002Fgrowth-log.md`** — operational state. Live thread state, drafts, pending replies, candidate handles in operational form. Gitignored. Updated by Oak and by `growth-watcher` \u002F `growth-day` skills.\n\nDo not write live operational state into `discovery-log.md`. Do not duplicate calibration lessons into `growth-log.md`. The boundary is sharp because the failure mode of conflating them is real — calibration evidence in the operational log produces noise; operational state in the calibration log produces leakage.\n\n## Out of scope\n\nThis skill does not:\n\n- Draft reply text or DM messages.\n- Plan engagement sequences (first \u002F second \u002F third touch).\n- Suggest off-platform contact paths.\n- Recommend follow-ups after silence.\n- Score posts for posting (that is `x-algorithm-grader`).\n- Generate posts (that is a future content skill).\n\nIf the user wants any of the above, return the list and surface the boundary plainly. The discovery output is the input to those skills, not their replacement.\n\n## Composing with other skills\n\n- `x-algorithm-grader` — for grading the user's *own* drafts (not for grading candidates' posts; cluster fit is a different question for posts vs people).\n- A future `x-engagement-planner` skill (not yet built) would consume the output of this skill and produce per-candidate first-touch sequences, second-touch contingencies, and reply-rate-aware cadence.\n\n## Reference files\n\n- `references\u002Faudience-profile.md` — the conversion criterion; who counts as a recruit, who does not, with one-line reasons for each excluded audience.\n- `references\u002Fsearch-strategy.md` — concrete `bird` CLI invocations, working vocabularies, and named anti-patterns.\n- `references\u002Fdiscovery-log.md` — empirical record of past discoveries, started from the @yieldfreaks \u002F @ta_eis_eauton \u002F @gingersamurai cases.\n\n## Why this skill exists\n\nTwo recent successful recruitments (`@yieldfreaks`, builder of an agent-registry health dashboard; `@ta_eis_eauton` \u002F Silverarrow, builder of an open-source agent-evaluation harness) shared a shape that was not visible until both had been found: small\u002Fmedium accounts shipping verifiable tooling adjacent to verification, evaluation, or registry infra in the OLAS \u002F Bittensor \u002F Numerai orbits. Without this skill the next discovery attempt would default to thesis-language matching, which earlier in the same session produced a list of enterprise CIOs, JS framework devs, and an OpenClaw agent. Capturing the working filter as a skill makes the next round repeatable and prevents the regression.\n",{},"next","f2aa6f30f2c7d7bcfc4c11e1f6b69f3f44c28603","growth\u002Farchive\u002Fskills\u002Fdiscover-twitter-recruits","https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono","https:\u002F\u002Fgithub.com\u002FJinn-Network","2026-11-06 19:00:02",{"source":7,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add Jinn-Network\u002Fmono --skill discover-twitter-recruits",{"tool":38,"command":39},"bunx","bunx skills add Jinn-Network\u002Fmono --skill discover-twitter-recruits",{"tool":41,"command":42},"pnpm","pnpm dlx skills add Jinn-Network\u002Fmono --skill discover-twitter-recruits","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Fgrowth\u002Farchive\u002Fskills\u002Fdiscover-twitter-recruits\u002FSKILL.md",true,[46,52,56],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Faudience-profile.md","REFERENCE","参考文档",9850,"https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Fgrowth\u002Farchive\u002Fskills\u002Fdiscover-twitter-recruits\u002Freferences\u002Faudience-profile.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fdiscovery-log.md",33544,"https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Fgrowth\u002Farchive\u002Fskills\u002Fdiscover-twitter-recruits\u002Freferences\u002Fdiscovery-log.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Fsearch-strategy.md",12703,"https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Fgrowth\u002Farchive\u002Fskills\u002Fdiscover-twitter-recruits\u002Freferences\u002Fsearch-strategy.md",[61,93,115,139,155,176],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":72,"publishedAt":21},"01M6HYJ7FA967P7Z0KJ98Y6EDB","jinn-operator","Reference for how a Jinn Network operator ran. The operator network is parked and is not running tasks, so use this to explain the jinn CLI, the MCP tools, and the Jinn protocol, not to set up, fund, or start an operator. Activates on mentions of \"jinn\", \"jinn agent\", \"jinn network\", \"jinn run\", \"Task\", \"SolverNet\", \"restoration\", or \"SolverPlugin\" in the context of operating an agent.","https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Foperator\u002Fskills\u002Fjinn-operator\u002FSKILL.md",[67],{"categoryNo":68,"categoryName":69,"taxonomyType":70,"taxonomyTypeDesc":71},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92],"operator","jinn operator","MCP","cli","tasks","run","start","network","SET","Tools","how","Bash","read","write","edit","context","use","up","operating","for",{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"publishedAt":21},"01M6HYJ34J55KTVBX4ECARYPYG","swe-rebench-v2-task","Reference for swe-rebench-v2.v1 task structure — input fields, repo setup, FAIL_TO_PASS\u002FPASS_TO_PASS semantics, the swe-rebench-v2-solution.v1 output schema, and how to submit a typed payload. Consult this skill when orienting on a task or constructing a solution.","https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Foperator\u002Fplugins\u002Fswe-rebench-v2-runtime\u002Fskills\u002Ftask\u002FSKILL.md",[],[100,101,102,83,103,104,105,106,107,108,109,110,111,112,113,114,92],"task","swe rebench v2 task","setup","consult","repo","schema","payload","reference","structure","input","submit","how to","repo setup","output","v1",{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":119,"categories":120,"keywords":124,"publishedAt":21},"01M6HYHZF9Z8857QWYNJX2HPAP","swe-rebench-v2-test-map","Walk PASS_TO_PASS test names to their source files via repo grep, identify the test-to-source ratio, and pre-load the relevant call graph into context before writing the patch.","https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Foperator\u002Fplugins\u002Fswe-rebench-v2-diffmin\u002Fskills\u002Ftest-map\u002FSKILL.md",[121],{"categoryNo":122,"categoryName":123,"taxonomyType":70,"taxonomyTypeDesc":71},"01JBSEEDCAT000000000000008","测试与质量",[125,126,127,128,129,130,104,88,131,132,133,134,135,136,137,138],"test","map","swe rebench v2 test map","writing","graph","grep","call graph","patch","files","call","via","relevant","load","source",{"skillNo":140,"name":141,"nameZh":6,"description":142,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":143,"categories":144,"keywords":145,"publishedAt":21},"01M6HYHW5ZD1JW791VGWT08KCG","swe-rebench-v2-diffmin","Bias the patch toward the smallest change that flips FAIL_TO_PASS without disturbing PASS_TO_PASS. Use diff_stats to validate hunk count, file count, and rename absence before submitting.","https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Foperator\u002Fplugins\u002Fswe-rebench-v2-diffmin\u002Fskills\u002Fdiffmin\u002FSKILL.md",[],[146,147,148,149,150,151,89,152,132,153,154],"swe rebench v2 diffmin","change","validate","file","stats","hunk","diff","rename","bias",{"skillNo":156,"name":157,"nameZh":6,"description":158,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":159,"categories":160,"keywords":161,"publishedAt":21},"01M6HYHR8FC1VWRCWC03B285HQ","learn","Use when running a goal end-to-end through a seven-phase learning loop. Sequences Orient → Strategize → Plan → Execute → Debrief → Improve → Memory consolidation, dispatching specialized subagents per phase, and improves itself between runs by mutating its own state directory.","https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Foperator\u002Fplugins\u002Flearner\u002Fskills\u002Flearn\u002FSKILL.md",[],[157,162,163,164,84,165,166,167,168,169,85,86,89,170,100,171,172,173,174,175],"directory","memory","state","learning","improve","loop","plan","specialized","goal","orient","end","strategize","debrief","consolidation",{"skillNo":177,"name":178,"nameZh":6,"description":179,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":180,"categories":181,"keywords":182,"publishedAt":21},"01M6HYHJNRRRFAFQ97Y2DQMB9S","jinn-repo-task","Reference for jinn-repo.v1 task structure — input fields, repo setup (checkout of Jinn-Network\u002Fmono at base_commit), the jinn-repo-solution.v1 output schema, and how the daemon harvests your diff. Consult this skill when orienting on a task or constructing a solution.","https:\u002F\u002Fgithub.com\u002FJinn-Network\u002Fmono\u002Fblob\u002Fnext\u002Foperator\u002Fplugins\u002Fjinn-repo-runtime\u002Fskills\u002Ftask\u002FSKILL.md",[],[104,100,183,184,80,83,103,152,105,185,186,108,109,187,112,113,188,189,114,92],"jinn repo task","commit","base","mono","checkout","daemon","at"]