[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77MT05ST98MWC204WJMB9TY":3,"skill-same:01M77MT05ST98MWC204WJMB9TY":55},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":28,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M77MT05ST98MWC204WJMB9TY","hive-adversarial-improve-current-project",null,"Load when hardening a specific component\u002Fpipeline of THIS project (your own src\u002F, tests\u002F, tasks\u002F — not the hivemind framework) by stress-testing it against difficult and hostile inputs — especially a trust boundary where untrusted content crosses into trusted execution (auth\u002Fvalidation seams, installers, parsers of fetched or user-supplied content). Triggers on \"hive-adversarial-improve-current-project this\", \"harden this component\", \"stress-test this boundary\", \"find failure modes in \u003Ccomponent>\".","HexingBot\u002Fhivemind","HexingBot","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F296674764?v=4","01M77MMXMW5SWSXKHB0VA4N1FF","01M77MMXNV3PZA3795D6PYX5C9",1,"MIT","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Fskills\u002Fhive-adversarial-improve-current-project\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: hive-adversarial-improve-current-project\ndescription: Load when hardening a specific component\u002Fpipeline of THIS project (your own src\u002F, tests\u002F, tasks\u002F — not the hivemind framework) by stress-testing it against difficult and hostile inputs — especially a trust boundary where untrusted content crosses into trusted execution (auth\u002Fvalidation seams, installers, parsers of fetched or user-supplied content). Triggers on \"hive-adversarial-improve-current-project this\", \"harden this component\", \"stress-test this boundary\", \"find failure modes in \u003Ccomponent>\".\n---\n\n# Hive Adversarial Improve (current project)\n\n**Context guard — consumer project only.** This skill targets **this project's own** components\n(its `src\u002F`, `tests\u002F`, `tasks\u002F`) — never the hivemind framework's internals. Before proceeding,\nconfirm you are NOT in the framework repo: call `isFrameworkRepo({ repoRoot })` from\n`src\u002Fframework-context.js` against the current working repo root, if that module is reachable in\nthis project. If it returns **true**, **STOP** — do not proceed with this skill — and direct the\nuser to the plain `hive-adversarial-improve` variant instead, which targets the framework's own\ninternals. If `src\u002Fframework-context.js` is not importable (the common case — most consumer projects\ndo not vendor the framework's source), that itself is evidence you are in a consumer project:\nproceed. Only STOP when you can positively confirm framework-repo identity (a\n`.claude-plugin\u002Fplugin.json` whose `name` is `hivemind`, AND a `src\u002F` directory, AND a `bin\u002Finit.js`\nfile, all present at the repo root).\n\nThis skill turns an ad-hoc \"let's think about how this could break\" conversation into a\nrepeatable, evidence-producing exercise. It is a **durable self-hardening capability**, not a\none-off brainstorm: every run is seeded from a cited failure-mode catalog, exercises the real\nshipped code, and converts every gap into a replayable regression fixture so the finding survives\nafter the session ends.\n\n## When to load this skill\n\nLoad whenever the task is to harden one specific, already-shipped component or pipeline **in this\nproject** by stress-testing it — especially at a **trust boundary** where content originating\noutside the project (a third-party input, a fetched URL, an uploaded file, an installer input,\nuser-supplied data) crosses into code that executes with the project's own privileges. Examples:\nhardening a form-input handler, stress-testing an auth pipeline, probing a config parser. Do not\nload it for the hivemind framework's own internals (that is `hive-adversarial-improve`; see the\ncontext guard above), for routine feature review (that's the `reviewer` agent), or for open-ended\nsecurity research with no concrete component in scope (that's the `researcher` agent).\n\n## The two load-bearing rules\n\nEverything below is scaffolding around these two rules. Drop either one and the exercise stops\nproducing durable value:\n\n1. **The live pipeline is the REAL shipped code — never a tabletop description.** The exercise must\n   execute the actual gate code, the actual CLI calls, the actual subagent spawns that ship in this\n   project today. A review where the pipeline is played \"in the abstract\" (a human narrating what the\n   code *would* probably do) is not a hardening review — it is a brainstorm with role names, and it\n   systematically misses the exact class of bug (a missed `await`, a regex that doesn't anchor, a\n   gate that's wired to the wrong function) that only shows up when the real code runs.\n2. **Every gap becomes a `tests-after` ticket, with the probe input as a replayable test fixture.** A\n   finding that lives only in a chat transcript evaporates the moment the session ends. The\n   acceptance criterion for that ticket is literally \"replaying this exact input is now caught\" —\n   the probe artifact itself is the fixture the new test asserts against.\n\n## The 9-step protocol\n\n1. **Name the component and its trust boundaries.** Write down every place untrusted content\n   crosses into trusted execution for this component (a fetched file being parsed, user-supplied\n   input being read by an agent or handler, a config value being interpolated into a shell call).\n   Each boundary gets at least one probe.\n2. **Assign roles.** The Reviewer (you, the orchestrator) runs the exercise. The Challenger is a\n   subagent briefed with the cited failure-mode catalog (`references\u002Ffailure-mode-catalog.md`) and\n   nothing else — no advance knowledge of the specific defenses beyond what an outside caller could\n   plausibly have. The live pipeline is the real shipped code, run for real (see rule 1 — this is\n   the single most important role assignment in the whole protocol). An optional Recorder takes\n   notes and tracks the round log without participating.\n3. **Seed one probe per trust boundary from a cited catalog.** Pull concrete probe ideas from\n   `references\u002Ffailure-mode-catalog.md` — do not improvise novel failure classes from first\n   principles. Improvised probes test the Challenger's imagination, not the component's real\n   exposure to the input classes that are actually seen in the wild.\n4. **Run rounds.** The Challenger proposes one concrete artifact (a crafted config value, a hostile\n   fetched response, a malformed user input). The live pipeline processes it through the actual code\n   path — the real gate, the real CLI, the real subagent. The Reviewer adjudicates the outcome as\n   caught \u002F missed \u002F partial and records **which gate** (by name — a specific function, regex, or\n   review step) was responsible for the catch or the miss.\n5. **Adapt.** After each round, the Challenger tries to get past whichever specific gate caught it\n   last round (a new alias, an extra indirection, a restructured input) — this is what surfaces the\n   difference between a check that generalizes and one that only catches the literal input you\n   started with.\n6. **Stop condition.** End the exercise when any of: a fixed round budget (N rounds) is spent, the\n   seeded catalog is exhausted, or a time-box expires. Fix the stop condition before round 1, not\n   after a convenient result appears.\n7. **Convert every gap into a `tests-after` ticket.** The ticket's acceptance criterion is \"replaying this\n   exact input is now caught,\" and the probe artifact produced in step 4 is attached as the test\n   fixture the new regression test asserts against (see rule 2).\n8. **Feed the round log back as detection signal.** New patterns discovered in step 5 become new\n   scanner rules and fixtures in this project's own detectors — whatever this project's analogous\n   input-validation layer is. This closes the loop from \"we found it once\" to \"we can no longer be\n   surprised by it.\"\n9. **Re-run the same probe set as a regression pass after fixes land.** Once every gap ticket from\n   step 7 is closed, replay the full original probe set (not a fresh one) against the fixed pipeline\n   to confirm every previously-missed input is now caught, and that no fix regressed a\n   previously-caught one.\n\n## References\n\n- `references\u002Ffailure-mode-catalog.md` — the supply-chain \u002F third-party-content input taxonomy, with\n  cited source URLs. Seed step 3's probes from here, not from imagination.\n- `references\u002Fuseless-vs-valuable.md` — the checklist for telling a hardening review that produced\n  durable value from one that produced only a transcript. Run through it before step 6's stop\n  condition and again before reporting results.\n\n## Severity scoring\n\nAdjudication in step 4 (and the tickets produced in step 7) reuse this project's reviewer\nHIGH \u002F MEDIUM \u002F LOW severity convention (the same three-tier scale the `reviewer` agent shipped with\nthe hivemind plugin already runs on) rather than importing a numeric rubric from the security-testing\nliterature — that literature is qualitative, and there is no cross-project standard scale worth\nadopting instead of the one already in use.\n\n## Worked example\n\nRead the `hive-adversarial-improve` skill's own \"Worked example\" section (framework repo only) for a\nconcrete reference run of this method end to end (`TASK-140` through `TASK-144` in the hivemind\nframework's own history) if the steps above feel abstract — the mechanic is identical here, only the\ntarget project differs: the live pipeline must always be this project's own real shipped code, run\nend to end, never a tabletop description of what it \"should\" do.\n","# Hive Adversarial Improve (current project)\n\n**Context guard — consumer project only.** This skill targets **this project's own** components\n(its `src\u002F`, `tests\u002F`, `tasks\u002F`) — never the hivemind framework's internals. Before proceeding,\nconfirm you are NOT in the framework repo: call `isFrameworkRepo({ repoRoot })` from\n`src\u002Fframework-context.js` against the current working repo root, if that module is reachable in\nthis project. If it returns **true**, **STOP** — do not proceed with this skill — and direct the\nuser to the plain `hive-adversarial-improve` variant instead, which targets the framework's own\ninternals. If `src\u002Fframework-context.js` is not importable (the common case — most consumer projects\ndo not vendor the framework's source), that itself is evidence you are in a consumer project:\nproceed. Only STOP when you can positively confirm framework-repo identity (a\n`.claude-plugin\u002Fplugin.json` whose `name` is `hivemind`, AND a `src\u002F` directory, AND a `bin\u002Finit.js`\nfile, all present at the repo root).\n\nThis skill turns an ad-hoc \"let's think about how this could break\" conversation into a\nrepeatable, evidence-producing exercise. It is a **durable self-hardening capability**, not a\none-off brainstorm: every run is seeded from a cited failure-mode catalog, exercises the real\nshipped code, and converts every gap into a replayable regression fixture so the finding survives\nafter the session ends.\n\n## When to load this skill\n\nLoad whenever the task is to harden one specific, already-shipped component or pipeline **in this\nproject** by stress-testing it — especially at a **trust boundary** where content originating\noutside the project (a third-party input, a fetched URL, an uploaded file, an installer input,\nuser-supplied data) crosses into code that executes with the project's own privileges. Examples:\nhardening a form-input handler, stress-testing an auth pipeline, probing a config parser. Do not\nload it for the hivemind framework's own internals (that is `hive-adversarial-improve`; see the\ncontext guard above), for routine feature review (that's the `reviewer` agent), or for open-ended\nsecurity research with no concrete component in scope (that's the `researcher` agent).\n\n## The two load-bearing rules\n\nEverything below is scaffolding around these two rules. Drop either one and the exercise stops\nproducing durable value:\n\n1. **The live pipeline is the REAL shipped code — never a tabletop description.** The exercise must\n   execute the actual gate code, the actual CLI calls, the actual subagent spawns that ship in this\n   project today. A review where the pipeline is played \"in the abstract\" (a human narrating what the\n   code *would* probably do) is not a hardening review — it is a brainstorm with role names, and it\n   systematically misses the exact class of bug (a missed `await`, a regex that doesn't anchor, a\n   gate that's wired to the wrong function) that only shows up when the real code runs.\n2. **Every gap becomes a `tests-after` ticket, with the probe input as a replayable test fixture.** A\n   finding that lives only in a chat transcript evaporates the moment the session ends. The\n   acceptance criterion for that ticket is literally \"replaying this exact input is now caught\" —\n   the probe artifact itself is the fixture the new test asserts against.\n\n## The 9-step protocol\n\n1. **Name the component and its trust boundaries.** Write down every place untrusted content\n   crosses into trusted execution for this component (a fetched file being parsed, user-supplied\n   input being read by an agent or handler, a config value being interpolated into a shell call).\n   Each boundary gets at least one probe.\n2. **Assign roles.** The Reviewer (you, the orchestrator) runs the exercise. The Challenger is a\n   subagent briefed with the cited failure-mode catalog (`references\u002Ffailure-mode-catalog.md`) and\n   nothing else — no advance knowledge of the specific defenses beyond what an outside caller could\n   plausibly have. The live pipeline is the real shipped code, run for real (see rule 1 — this is\n   the single most important role assignment in the whole protocol). An optional Recorder takes\n   notes and tracks the round log without participating.\n3. **Seed one probe per trust boundary from a cited catalog.** Pull concrete probe ideas from\n   `references\u002Ffailure-mode-catalog.md` — do not improvise novel failure classes from first\n   principles. Improvised probes test the Challenger's imagination, not the component's real\n   exposure to the input classes that are actually seen in the wild.\n4. **Run rounds.** The Challenger proposes one concrete artifact (a crafted config value, a hostile\n   fetched response, a malformed user input). The live pipeline processes it through the actual code\n   path — the real gate, the real CLI, the real subagent. The Reviewer adjudicates the outcome as\n   caught \u002F missed \u002F partial and records **which gate** (by name — a specific function, regex, or\n   review step) was responsible for the catch or the miss.\n5. **Adapt.** After each round, the Challenger tries to get past whichever specific gate caught it\n   last round (a new alias, an extra indirection, a restructured input) — this is what surfaces the\n   difference between a check that generalizes and one that only catches the literal input you\n   started with.\n6. **Stop condition.** End the exercise when any of: a fixed round budget (N rounds) is spent, the\n   seeded catalog is exhausted, or a time-box expires. Fix the stop condition before round 1, not\n   after a convenient result appears.\n7. **Convert every gap into a `tests-after` ticket.** The ticket's acceptance criterion is \"replaying this\n   exact input is now caught,\" and the probe artifact produced in step 4 is attached as the test\n   fixture the new regression test asserts against (see rule 2).\n8. **Feed the round log back as detection signal.** New patterns discovered in step 5 become new\n   scanner rules and fixtures in this project's own detectors — whatever this project's analogous\n   input-validation layer is. This closes the loop from \"we found it once\" to \"we can no longer be\n   surprised by it.\"\n9. **Re-run the same probe set as a regression pass after fixes land.** Once every gap ticket from\n   step 7 is closed, replay the full original probe set (not a fresh one) against the fixed pipeline\n   to confirm every previously-missed input is now caught, and that no fix regressed a\n   previously-caught one.\n\n## References\n\n- `references\u002Ffailure-mode-catalog.md` — the supply-chain \u002F third-party-content input taxonomy, with\n  cited source URLs. Seed step 3's probes from here, not from imagination.\n- `references\u002Fuseless-vs-valuable.md` — the checklist for telling a hardening review that produced\n  durable value from one that produced only a transcript. Run through it before step 6's stop\n  condition and again before reporting results.\n\n## Severity scoring\n\nAdjudication in step 4 (and the tickets produced in step 7) reuse this project's reviewer\nHIGH \u002F MEDIUM \u002F LOW severity convention (the same three-tier scale the `reviewer` agent shipped with\nthe hivemind plugin already runs on) rather than importing a numeric rubric from the security-testing\nliterature — that literature is qualitative, and there is no cross-project standard scale worth\nadopting instead of the one already in use.\n\n## Worked example\n\nRead the `hive-adversarial-improve` skill's own \"Worked example\" section (framework repo only) for a\nconcrete reference run of this method end to end (`TASK-140` through `TASK-144` in the hivemind\nframework's own history) if the steps above feel abstract — the mechanic is identical here, only the\ntarget project differs: the live pipeline must always be this project's own real shipped code, run\nend to end, never a tabletop description of what it \"should\" do.\n",{"name":5,"description":7},"main","c8856147b7a3246a1e54b66126d093183805d1f3","skills\u002Fhive-adversarial-improve-current-project",0,"https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind","https:\u002F\u002Fhivemind-ef84a.web.app","https:\u002F\u002Fgithub.com\u002FHexingBot","2026-10-10 00:50:01",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add HexingBot\u002Fhivemind --skill hive-adversarial-improve-current-project",{"tool":37,"command":38},"bunx","bunx skills add HexingBot\u002Fhivemind --skill hive-adversarial-improve-current-project",{"tool":40,"command":41},"pnpm","pnpm dlx skills add HexingBot\u002Fhivemind --skill hive-adversarial-improve-current-project","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Fskills\u002Fhive-adversarial-improve-current-project\u002FSKILL.md",true,[45,51],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"references\u002Ffailure-mode-catalog.md","REFERENCE","参考文档",4772,"https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Fskills\u002Fhive-adversarial-improve-current-project\u002Freferences\u002Ffailure-mode-catalog.md",{"path":52,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":53,"sourceHtmlUrl":54},"references\u002Fuseless-vs-valuable.md",3598,"https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Fskills\u002Fhive-adversarial-improve-current-project\u002Freferences\u002Fuseless-vs-valuable.md",[56,63,72,80,87,95],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":18},"01M77MWBWY2AK0GJ91MMKAPZ0M","unknown-skill","A demo skill with no license signal at all, used to exercise the \"unknown\" classification branch of the human license gate (TASK-135).","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Funknown-skill\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":67,"licenseFileUrl":68,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":18},"01M77MW9H43Z0ATZ15421K8J7D","spoofed-license-skill","A demo skill used to test TASK-150's SPDX-header-vs-LICENSE-file conflict surfacing.","GPL-3.0","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fspoofed-license-skill\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fspoofed-license-skill\u002FSKILL.md",[],[],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":76,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":18},"01M77MW7715ENRZN6F01FDM81J","permissive-skill","A demo skill used to test assimilation of a permissively-licensed third party skill.","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fpermissive-skill\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fpermissive-skill\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":18},"01M77MW3CRAPWETDAZT7E08BVS","malicious-skill","A fixture skill deliberately carrying risky content patterns, used only to exercise src\u002Fskill-scan.js (TASK-122). Never assimilated for real.","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fmalicious-skill\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":67,"licenseFileUrl":91,"sourceHtmlUrl":92,"categories":93,"keywords":94,"publishedAt":18},"01M77MW1RSWW1MTQGM9TY3X37G","copyleft-skill","A demo skill used to test the human license gate for a copyleft third party skill.","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fcopyleft-skill\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fcopyleft-skill\u002FSKILL.md",[],[],{"skillNo":96,"name":97,"nameZh":6,"description":98,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":101,"publishedAt":18},"01M77MVWT1EVEJ8ZYZTCG4CENE","watch","Watch a video (URL or local path). Downloads with yt-dlp, extracts auto-scaled frames with ffmpeg, pulls the transcript from captions (or Whisper API fallback), and hands the result to Claude so it can answer questions about what's in the video. (assimilated for Hivemind)","https:\u002F\u002Fgithub.com\u002FHexingBot\u002Fhivemind\u002Fblob\u002Fmain\u002Fskills\u002Fwatch\u002FSKILL.md",[],[]]