[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6JGHTQM6RW85V8XNF38HYD9":3,"skill-same:01M6JGHTQM6RW85V8XNF38HYD9":139},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":39,"linkOnly":40,"markdownContent":41,"markdownBody":42,"frontMatter":43,"refName":62,"commitSha":63,"skillDirectory":64,"forksCount":65,"repositoryHtmlUrl":66,"repositoryHomepageUrl":6,"ownerHtmlUrl":67,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":68,"install":69,"resources":82},"01M6JGHTQM6RW85V8XNF38HYD9","safety-guard",null,"PreToolUse hooks that refuse destructive commands, ask before a shared-state write, block raw merges of task branches into a base branch outside the review flow, refuse a gate flag the agent writes for itself, keep the tracker writable only through `rack`, refuse the harness task list, and turn a consent-declared move into a question for the supervisor. Infrastructure — read it to diagnose a block, not to invoke it.","muratovv\u002Fai-hats","muratovv","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F7502801?v=4","01M6JGCBN4KEJX012GPJVSH04T","Fedor Muratov","01M6JGCBSNMN7CW3RGTPPE0MN2",3,"MIT","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38],"Safety","guard","safety guard","only","for","move","keep","shared","backlog","branches","task","hooks","review","question","ask","branch","supervisor","Bash","harness","diagnose","2026-11-10 10:12:30",false,"---\nname: safety-guard\ndescription: >-\n  PreToolUse hooks that refuse destructive commands, ask before a\n  shared-state write, block raw merges of task branches into a base branch\n  outside the review flow, refuse a gate flag the agent writes for itself,\n  keep the tracker writable only through `rack`, refuse the harness task\n  list, and turn a consent-declared move into a question for the\n  supervisor. Infrastructure — read it to diagnose a block, not to invoke\n  it.\nai_hats:\n  runtime_hooks:\n    PreToolUse:\n      - matcher: Bash|run_command|execute\n        script: hooks\u002Fsafety_gate.py\n      - matcher: Bash\n        script: hooks\u002Fpre_bash_shared_state_guard.sh\n      - matcher: Bash\n        script: hooks\u002Fpre_bash_task_merge_guard.sh\n      - matcher: Bash\n        script: hooks\u002Fack_prefix_guard.py\n      - matcher: Edit|Write|MultiEdit\n        script: hooks\u002Fbacklog_write_gate.py\n      - matcher: TaskCreate|TaskUpdate|TaskList|TaskGet|TaskStop|TaskOutput\n        script: hooks\u002Fharness_task_gate.py\nlicense: MIT\n---\n\n# safety-guard\n\nA PreToolUse runtime-hook for Bash\u002Fterminal tools. It parses the command line\n(quote-aware) and matches **what the command targets**, not what it is named,\nbecause `global_rule_destructive_actions` protects paths.\n\n`PreToolUse` is the Claude surface's name for the point; each other surface\npackage carries its own dispatcher or adapter for it. What is NOT uniform is\nthe **dialect of the refusal**: a caller whose payload declares `PreToolUse`\ngets a decision object and the user is asked, while every other caller gets\n`exit 2` with the reason on stderr — a hard block, with no one to ask. So the\nsame command can be a question on one surface and a refusal on another, and\nneither outcome is the guard failing.\n\n## The three outcomes\n\n| Target                                                                                                                                                                                            | Outcome                                      |\n| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------- |\n| Scratch space, build artefacts, `\u002Ftmp` — the cleanup `global_rule_resource_hygiene` mandates                                                                                                      | allowed                                      |\n| Protected data — anything not regenerable from repo + toolchain; matched by the fast path `*.db` `*.sqlite*` `*.sql` `*.dump`, `volumes\u002F` `data\u002F` `storage\u002F` `runs\u002F`, `terraform.tfstate`, `.env` | denied; `AI_HATS_DESTRUCTIVE_ACK=1` opens it |\n| Filesystem root, `$HOME`, disk-formatting (`mkfs.*`), `dd of=\u002Fdev\u002F…`                                                                                                                              | denied; no flag opens it                     |\n\nEvery denial names the flag that unblocks it, or says plainly that none does.\nA guard that can only say \"no\" pushes the agent toward blunt instruments.\n\nEvery flag named here is read from the environment that **launched** the agent.\nA hook runs before the command it judges is a process, so a prefix on that command\nreaches nothing — the hint that once advertised one was measured denying the very\nline it told the agent to write.\n\n`git push` and a reversible shared write (`gh pr create`, `gh issue comment`,\n`gh release create`) are **not** handled here — `pre_bash_shared_state_guard.sh`\nasks before each of them. Two gates on one concern means the coarser one\nsilently wins.\n\n## An approval you write yourself is not one\n\n`ack_prefix_guard.py` refuses a Bash line that SETS an `AI_HATS_*` approval —\n`AI_HATS_SOME_GATE_ACK=1 make done-gate`, the bare assignment, `export`, `env`,\nand the same after a separator.\n\nThe gap it closes was never a policy, only a process boundary. A flag read by a\nPreToolUse guard comes from the environment that LAUNCHED the agent, so a prefix\ncannot reach it. A flag read by a script inside `make` or a git hook is reached by\nplain shell semantics — so the hatch the `master-ci` stage once carried was\nself-servable while the others were not, and the smoke gate's was skipped on four\ncommits that way.\n\nIt recognises the flag by SHAPE — the `_ACK` \u002F `_OFF` \u002F `_SKIP` spelling that\n`constants.withheld_from_subagent` uses — never by a roster, so a flag added next\nmonth is covered without anyone remembering. What it does **not** touch: the flag\nin argument position. `grep -rn AI_HATS_SOME_GATE_ACK scripts\u002F` and an `echo` of\nthe same text bind nothing and pass, which is the whole precision of the thing.\n\n**There is no flag that opens this guard**, by construction: one would be the same\nself-grant a level up. The way through is the supervisor setting the flag in the\nlaunching environment — and where it is then honoured, the use is journalled.\n\nConsent's own spellings are excluded here; `safety_gate.py` owns those, for the\nreason `git push` is owned elsewhere.\n\n## Some moves are consent-gated\n\nOn a few moves the guard turns your command into a question for the supervisor\ninstead of running it. Which moves those are is the role's declaration, not\nyours to know here. Three things you can do, and nothing else:\n\n- **Run the command bare and let the question happen.** Do not prefix it with a\n  consent flag or a ticket — the guard mints those, and one you typed is refused\n  as a self-grant.\n- **Present what you are asking approval for, and stop.** While a question is\n  waiting, do not nudge or re-ask: it does not expire, and the supervisor may\n  take as long as reading needs.\n- **If the engine refused for want of an answer, say so and stop.** A re-run\n  helps only where a question can be raised at all; on a surface with no hooks\n  it buys a second refusal. What opens the move is the supervisor typing the\n  verb himself (`consent \u003Ctype> \u003Cminutes>`), and his window then covers the\n  series without asking again. Mutating the command never helps: the answer\n  binds to the exact call he was shown, and a consent flag you add yourself is\n  a self-grant.\n\nA **critical-path removal** is asked every time, and no window covers it. That\nis an `rm` \u002F `rmdir` of the working directory or an ancestor, a top-level\ndirectory, or an operand holding `$` or a backtick. Where the removal can be\nnarrowed, narrow it: name the literal path under the working directory rather\nthan `\"$DIR\"\u002F*`. Where it cannot, run it bare and wait for the answer.\n\n## The harness task list is `rack`-only\n\n`harness_task_gate.py` denies `TaskCreate`, `TaskUpdate`, `TaskList`, `TaskGet`,\n`TaskStop` and `TaskOutput`. The refusal names `rack`. A harness without those\ntools never matches, so the hook stays quiet. The supervisor exports\n`AI_HATS_HARNESS_TASKS_OFF=1` for a session that has no backlog CLI.\n\n## The tracker backlog is `rack`-only\n\n`backlog_write_gate.py` (Edit\u002FWrite\u002FMultiEdit) and a matching predicate in\n`safety_gate.py` (Bash) deny writes under `\u003Cai_hats_dir>\u002Ftracker\u002Fbacklog\u002F**`.\nA hand-edited card desynchronises the state\nmachine from its locks and audit trail, so the card moves with\n`rack transition \u003CID> \u003Cstate> --log \"...\"`, fields change with\n`rack transition \u003CID> --set \u003Cfield>=\u003Cvalue>`, and a document (`summary.md`,\n`audit.md`, …) is written outside the tracker and brought in with\n`rack transition \u003CID> --attach \u002Ftmp\u002Fsummary.md:summary.md`. Reads (`cat`,\n`grep`) are not touched, and `tasks\u002F\u003CID>\u002Fplan.md` stays directly writable\n— it is the agent's own deliverable, not FSM-owned state.\n\n`ai_hats_dir` is resolved from the TARGET path's own `ai-hats.yaml`, never from\n`$AI_HATS_DIR`: that variable leaks between checkouts, and a worktree session\nediting the main checkout's tracker must be judged by that tracker's config.\n\nThere is no per-call flag. If the tracker is broken and only a raw edit can\nrepair it, the supervisor exports `AI_HATS_BACKLOG_GATE_OFF=1` for the session.\n\n## YOLO Mode\n\nIf you expect to run a large number of destructive commands (e.g. bulk file deletion, complex refactors requiring many in-place edits), you can request YOLO mode from the user to bypass the safety gate for the duration of your session.\nTo do this, attempt to execute `export AI_HATS_YOLO=1` in your Bash tool. The safety gate will intercept this and prompt the supervisor for permission.\n\n**Export it; do not prefix it.** The gate reads `AI_HATS_YOLO` from its own\nenvironment, so `AI_HATS_YOLO=1 \u003Ccommand>` inline is refused by design: a guard\nthe agent can switch off per-command is not a guard. Only an exported value,\nwhich the supervisor's shell grants, takes effect.\n\n**CRITICAL RULES FOR YOLO MODE:**\n\n1. You may request YOLO mode from the supervisor.\n2. You MUST NOT request YOLO mode arbitrarily or \"just in case\".\n3. When requesting YOLO mode (which triggers a prompt to the supervisor), you MUST explicitly output a message stating *why* you need it and exactly what destructive actions you plan to perform.\n","# safety-guard\n\nA PreToolUse runtime-hook for Bash\u002Fterminal tools. It parses the command line\n(quote-aware) and matches **what the command targets**, not what it is named,\nbecause `global_rule_destructive_actions` protects paths.\n\n`PreToolUse` is the Claude surface's name for the point; each other surface\npackage carries its own dispatcher or adapter for it. What is NOT uniform is\nthe **dialect of the refusal**: a caller whose payload declares `PreToolUse`\ngets a decision object and the user is asked, while every other caller gets\n`exit 2` with the reason on stderr — a hard block, with no one to ask. So the\nsame command can be a question on one surface and a refusal on another, and\nneither outcome is the guard failing.\n\n## The three outcomes\n\n| Target                                                                                                                                                                                            | Outcome                                      |\n| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------- |\n| Scratch space, build artefacts, `\u002Ftmp` — the cleanup `global_rule_resource_hygiene` mandates                                                                                                      | allowed                                      |\n| Protected data — anything not regenerable from repo + toolchain; matched by the fast path `*.db` `*.sqlite*` `*.sql` `*.dump`, `volumes\u002F` `data\u002F` `storage\u002F` `runs\u002F`, `terraform.tfstate`, `.env` | denied; `AI_HATS_DESTRUCTIVE_ACK=1` opens it |\n| Filesystem root, `$HOME`, disk-formatting (`mkfs.*`), `dd of=\u002Fdev\u002F…`                                                                                                                              | denied; no flag opens it                     |\n\nEvery denial names the flag that unblocks it, or says plainly that none does.\nA guard that can only say \"no\" pushes the agent toward blunt instruments.\n\nEvery flag named here is read from the environment that **launched** the agent.\nA hook runs before the command it judges is a process, so a prefix on that command\nreaches nothing — the hint that once advertised one was measured denying the very\nline it told the agent to write.\n\n`git push` and a reversible shared write (`gh pr create`, `gh issue comment`,\n`gh release create`) are **not** handled here — `pre_bash_shared_state_guard.sh`\nasks before each of them. Two gates on one concern means the coarser one\nsilently wins.\n\n## An approval you write yourself is not one\n\n`ack_prefix_guard.py` refuses a Bash line that SETS an `AI_HATS_*` approval —\n`AI_HATS_SOME_GATE_ACK=1 make done-gate`, the bare assignment, `export`, `env`,\nand the same after a separator.\n\nThe gap it closes was never a policy, only a process boundary. A flag read by a\nPreToolUse guard comes from the environment that LAUNCHED the agent, so a prefix\ncannot reach it. A flag read by a script inside `make` or a git hook is reached by\nplain shell semantics — so the hatch the `master-ci` stage once carried was\nself-servable while the others were not, and the smoke gate's was skipped on four\ncommits that way.\n\nIt recognises the flag by SHAPE — the `_ACK` \u002F `_OFF` \u002F `_SKIP` spelling that\n`constants.withheld_from_subagent` uses — never by a roster, so a flag added next\nmonth is covered without anyone remembering. What it does **not** touch: the flag\nin argument position. `grep -rn AI_HATS_SOME_GATE_ACK scripts\u002F` and an `echo` of\nthe same text bind nothing and pass, which is the whole precision of the thing.\n\n**There is no flag that opens this guard**, by construction: one would be the same\nself-grant a level up. The way through is the supervisor setting the flag in the\nlaunching environment — and where it is then honoured, the use is journalled.\n\nConsent's own spellings are excluded here; `safety_gate.py` owns those, for the\nreason `git push` is owned elsewhere.\n\n## Some moves are consent-gated\n\nOn a few moves the guard turns your command into a question for the supervisor\ninstead of running it. Which moves those are is the role's declaration, not\nyours to know here. Three things you can do, and nothing else:\n\n- **Run the command bare and let the question happen.** Do not prefix it with a\n  consent flag or a ticket — the guard mints those, and one you typed is refused\n  as a self-grant.\n- **Present what you are asking approval for, and stop.** While a question is\n  waiting, do not nudge or re-ask: it does not expire, and the supervisor may\n  take as long as reading needs.\n- **If the engine refused for want of an answer, say so and stop.** A re-run\n  helps only where a question can be raised at all; on a surface with no hooks\n  it buys a second refusal. What opens the move is the supervisor typing the\n  verb himself (`consent \u003Ctype> \u003Cminutes>`), and his window then covers the\n  series without asking again. Mutating the command never helps: the answer\n  binds to the exact call he was shown, and a consent flag you add yourself is\n  a self-grant.\n\nA **critical-path removal** is asked every time, and no window covers it. That\nis an `rm` \u002F `rmdir` of the working directory or an ancestor, a top-level\ndirectory, or an operand holding `$` or a backtick. Where the removal can be\nnarrowed, narrow it: name the literal path under the working directory rather\nthan `\"$DIR\"\u002F*`. Where it cannot, run it bare and wait for the answer.\n\n## The harness task list is `rack`-only\n\n`harness_task_gate.py` denies `TaskCreate`, `TaskUpdate`, `TaskList`, `TaskGet`,\n`TaskStop` and `TaskOutput`. The refusal names `rack`. A harness without those\ntools never matches, so the hook stays quiet. The supervisor exports\n`AI_HATS_HARNESS_TASKS_OFF=1` for a session that has no backlog CLI.\n\n## The tracker backlog is `rack`-only\n\n`backlog_write_gate.py` (Edit\u002FWrite\u002FMultiEdit) and a matching predicate in\n`safety_gate.py` (Bash) deny writes under `\u003Cai_hats_dir>\u002Ftracker\u002Fbacklog\u002F**`.\nA hand-edited card desynchronises the state\nmachine from its locks and audit trail, so the card moves with\n`rack transition \u003CID> \u003Cstate> --log \"...\"`, fields change with\n`rack transition \u003CID> --set \u003Cfield>=\u003Cvalue>`, and a document (`summary.md`,\n`audit.md`, …) is written outside the tracker and brought in with\n`rack transition \u003CID> --attach \u002Ftmp\u002Fsummary.md:summary.md`. Reads (`cat`,\n`grep`) are not touched, and `tasks\u002F\u003CID>\u002Fplan.md` stays directly writable\n— it is the agent's own deliverable, not FSM-owned state.\n\n`ai_hats_dir` is resolved from the TARGET path's own `ai-hats.yaml`, never from\n`$AI_HATS_DIR`: that variable leaks between checkouts, and a worktree session\nediting the main checkout's tracker must be judged by that tracker's config.\n\nThere is no per-call flag. If the tracker is broken and only a raw edit can\nrepair it, the supervisor exports `AI_HATS_BACKLOG_GATE_OFF=1` for the session.\n\n## YOLO Mode\n\nIf you expect to run a large number of destructive commands (e.g. bulk file deletion, complex refactors requiring many in-place edits), you can request YOLO mode from the user to bypass the safety gate for the duration of your session.\nTo do this, attempt to execute `export AI_HATS_YOLO=1` in your Bash tool. The safety gate will intercept this and prompt the supervisor for permission.\n\n**Export it; do not prefix it.** The gate reads `AI_HATS_YOLO` from its own\nenvironment, so `AI_HATS_YOLO=1 \u003Ccommand>` inline is refused by design: a guard\nthe agent can switch off per-command is not a guard. Only an exported value,\nwhich the supervisor's shell grants, takes effect.\n\n**CRITICAL RULES FOR YOLO MODE:**\n\n1. You may request YOLO mode from the supervisor.\n2. You MUST NOT request YOLO mode arbitrarily or \"just in case\".\n3. When requesting YOLO mode (which triggers a prompt to the supervisor), you MUST explicitly output a message stating *why* you need it and exactly what destructive actions you plan to perform.\n",{"name":5,"ai_hats":44,"license":15,"description":7},{"runtime_hooks":45},{"PreToolUse":46},[47,50,52,54,56,59],{"script":48,"matcher":49},"hooks\u002Fsafety_gate.py","Bash|run_command|execute",{"script":51,"matcher":36},"hooks\u002Fpre_bash_shared_state_guard.sh",{"script":53,"matcher":36},"hooks\u002Fpre_bash_task_merge_guard.sh",{"script":55,"matcher":36},"hooks\u002Fack_prefix_guard.py",{"script":57,"matcher":58},"hooks\u002Fbacklog_write_gate.py","Edit|Write|MultiEdit",{"script":60,"matcher":61},"hooks\u002Fharness_task_gate.py","TaskCreate|TaskUpdate|TaskList|TaskGet|TaskStop|TaskOutput","master","b4819add93efbaf4d80627118ec34680406b309b","packages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard",1,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats","https:\u002F\u002Fgithub.com\u002Fmuratovv","2026-11-05 20:59:10",{"source":8,"skillName":5,"commands":70,"prompt":80,"zipAvailable":81},[71,74,77],{"tool":72,"command":73},"npx","npx skills add muratovv\u002Fai-hats --skill safety-guard",{"tool":75,"command":76},"bunx","bunx skills add muratovv\u002Fai-hats --skill safety-guard",{"tool":78,"command":79},"pnpm","pnpm dlx skills add muratovv\u002Fai-hats --skill safety-guard","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002FSKILL.md",true,[83,88,91,95,98,104,108,111,114,118,121,124,127,131,135],{"path":55,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":86,"sourceHtmlUrl":87},"SCRIPT","脚本",5712,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fack_prefix_guard.py",{"path":57,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":89,"sourceHtmlUrl":90},9602,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fbacklog_write_gate.py",{"path":92,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":93,"sourceHtmlUrl":94},"hooks\u002Fbypass_journal.py",35,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fbypass_journal.py",{"path":96,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":93,"sourceHtmlUrl":97},"hooks\u002Fbypass_journal.sh","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fbypass_journal.sh",{"path":99,"resourceType":100,"resourceTypeDesc":101,"sizeBytes":102,"sourceHtmlUrl":103},"hooks\u002Fconsent_gate","OTHER","其他",30,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fconsent_gate",{"path":105,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":106,"sourceHtmlUrl":107},"hooks\u002Fconsent_spellings.py",3948,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fconsent_spellings.py",{"path":109,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":93,"sourceHtmlUrl":110},"hooks\u002Fconsent_ticket.py","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fconsent_ticket.py",{"path":60,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":112,"sourceHtmlUrl":113},2617,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fharness_task_gate.py",{"path":115,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":116,"sourceHtmlUrl":117},"hooks\u002Fmerge_order_check.sh",7170,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fmerge_order_check.sh",{"path":51,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":119,"sourceHtmlUrl":120},8954,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fpre_bash_shared_state_guard.sh",{"path":53,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":122,"sourceHtmlUrl":123},9063,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fpre_bash_task_merge_guard.sh",{"path":48,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":125,"sourceHtmlUrl":126},55465,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fsafety_gate.py",{"path":128,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":129,"sourceHtmlUrl":130},"hooks\u002Fshared_state_classifier.sh",5384,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fshared_state_classifier.sh",{"path":132,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":133,"sourceHtmlUrl":134},"hooks\u002Fshell_walk.py",31,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Fshell_walk.py",{"path":136,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":137,"sourceHtmlUrl":138},"hooks\u002Ftask_merge_classifier.sh",3956,"https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fcore\u002Fskills\u002Fsafety-guard\u002Fhooks\u002Ftask_merge_classifier.sh",[140,166,191,214,238,252],{"skillNo":141,"name":142,"nameZh":6,"description":143,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":144,"categories":145,"keywords":146,"publishedAt":39},"01M6PVFB1SM738R2SRSRSF7CY6","tool-evaluation-protocol","Time-bounded protocol for evaluating new tools, libraries, or methodologies before adoption. Use when evaluating a candidate dependency for the project, researching a methodology (e.g. ML\u002FLLM frameworks, testing approaches), or following up on a hype-driven recommendation.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Ftool-evaluation-protocol\u002FSKILL.md",[],[147,148,149,150,151,152,153,154,155,156,157,158,159,160,161,162,163,23,164,165],"protocol","Evaluation","tool","tool evaluation protocol","testing","ml","libraries","adoption","methodology","Tools","use","dependency","up","frameworks","time","project","recommendation","bounded","candidate",{"skillNo":167,"name":168,"nameZh":6,"description":169,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":170,"categories":171,"keywords":172,"publishedAt":39},"01M6PVF7HATBEX1G7N50E2QT57","systematic-debugging","4-phase bug-fix protocol (evidence, pattern, hypothesis, verify). Use when investigating any error, failure, or unexpected behavior, handling bug reports from users or monitoring, or diagnosing test failures with non-obvious causes.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fsystematic-debugging\u002FSKILL.md",[],[173,174,175,176,147,177,178,179,180,181,157,182,183,184,185,186,187,188,189,190],"systematic debugging","debugging","systematic","test","monitoring","hypothesis","verify","behavior","fix","evidence","bug","reports","pattern","users","error","with","bug-fix","phase",{"skillNo":192,"name":193,"nameZh":6,"description":194,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":195,"categories":196,"keywords":202,"publishedAt":39},"01M6PVF3MTQN3RPF73AN34856J","skill-template","Canonical template and validation guide for ai-hats skills. Use when creating a new skill, reviewing an existing one for structural compliance, or deciding which pattern fits a new behavior.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fskill-template\u002FSKILL.md",[197],{"categoryNo":198,"categoryName":199,"taxonomyType":200,"taxonomyTypeDesc":201},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[203,204,205,206,207,180,208,157,209,185,210,211,212,23,213],"template","skill template","validation","compliance","hats","guide","new skill","structural","canonical","which","Existing",{"skillNo":215,"name":216,"nameZh":6,"description":217,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":218,"categories":219,"keywords":221,"publishedAt":39},"01M6PVF06SCV28QARM5MQPANC9","skill-optimization","Audit and refactor library components to eliminate redundancy and staleness. Use during periodic library maintenance, after adding multiple new skills or rules, when a role's token budget is too high, or after retrospectives identify component issues.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fskill-optimization\u002FSKILL.md",[220],{"categoryNo":198,"categoryName":199,"taxonomyType":200,"taxonomyTypeDesc":201},[222,223,224,225,226,227,228,229,230,231,157,232,233,234,235,236,237],"Optimization","skill optimization","components","component","issues","library","audit","refactor","Budget","Rules","periodic","maintenance","token","token budget","role","high",{"skillNo":239,"name":240,"nameZh":6,"description":241,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":242,"categories":243,"keywords":244,"publishedAt":39},"01M6PVEXM9717016T9MW82HDCT","silent-fallback-gate","PreToolUse hook — refuses an Edit or Write that adds an inert broad except. Use when a Python edit was blocked for a silent fallback.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fsilent-fallback-gate\u002FSKILL.md",[],[245,246,247,248,249,250,157,251,23],"gate","fallback","silent fallback gate","python","write","edit","hook",{"skillNo":253,"name":254,"nameZh":6,"description":255,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":256,"categories":257,"keywords":259,"publishedAt":39},"01M6PVESM49C5NT7SB7N3F76VS","retro-to-framework","Convert an observation about agent behavior — from a retrospective or from a single incident — into a framework-level change (hook, rule, skill, skill update). Use after a retrospective identifies problems that are not project-specific, after an incident where an agent misapplied a component it had read, when CLAUDE.md band-aids accumulate (more than 3 per-project rules that could be generic), or when the same problem recurs across multiple projects.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fretro-to-framework\u002FSKILL.md",[258],{"categoryNo":198,"categoryName":199,"taxonomyType":200,"taxonomyTypeDesc":201},[260,261,262,263,225,264,265,266,231,180,157,267,268,162,269,270,271,272,273,274],"framework","retro","retro to framework","change","retrospective","projects","convert","incident","across","problem","about","more","skill update","single","generic"]