[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M758H72XMX10VGX2DSJKDGWM":3,"skill-same:01M758H72XMX10VGX2DSJKDGWM":68},{"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":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":31,"commitSha":32,"skillDirectory":33,"forksCount":34,"repositoryHtmlUrl":35,"repositoryHomepageUrl":36,"ownerHtmlUrl":37,"ownerBlogUrl":6,"ownerDescription":38,"updatedAt":39,"install":40,"resources":53},"01M758H72XMX10VGX2DSJKDGWM","skill-curation",null,"Consolidate, dedupe, and prune the skill library — merge overlapping skills into one (absorbed_into), delete truly dead skills, and keep the manifest honest. Use when the doctor flags skill drift, two skills overlap, or a lifecycle run surfaces consolidation candidates.","lukemcqueen\u002Fsteadfaste-fleet","lukemcqueen","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F405240?v=4","01M757MHPAGRXTJAT46SKZP98R","Luke McQueen","01M757MHPZ573KP8DT9GB8SBQ7",4,"MIT","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fskill-curation\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: skill-curation\nversion: 1.1.0\ndescription: \"Consolidate, dedupe, and prune the skill library — merge overlapping skills into one (absorbed_into), delete truly dead skills, and keep the manifest honest. Use when the doctor flags skill drift, two skills overlap, or a lifecycle run surfaces consolidation candidates.\"\ntriggers:\n  - \"curate skills\"\n  - \"consolidate skills\"\n  - \"merge skills\"\n  - \"prune skills\"\n  - \"skill dedupe\"\n---\n\n# Skill Curation\n\nConsolidate, dedupe, and prune the skill library. Curation is the\ncounterweight to creation: every new skill is debt until it earns its place,\nand a merged skill serves better than two overlapping ones.\n\n## When to Curate\n\n- After a lifecycle run (orch-skill-lifecycle Monday deep eval)\n- When 2+ skills overlap — `skills_list(category=...)` shows same-domain\n  entries with similar descriptions\n- On any deletion — always declare `absorbed_into=\u003Cumbrella>` (merged\n  into an existing skill) or `absorbed_into=\"\"` (true prune) so downstream\n  tooling (cron-job skill references, the curator) knows intent\n- When the doctor flags skill drift, duplicates, or deployed orphans\n\n## Rules\n\n1. **Prove duplication first.** Load both skills and compare structure AND\n   content — descriptions alone lie. Overlap > 50% → merge; otherwise keep\n   both and note the boundary.\n2. **Improve existing before creating.** The merge target is the existing\n   skill — extend it, never spawn a sibling. If the two are equal weight,\n   keep the one with better structure, fold the other's unique sections in.\n3. **Redirects carry real bodies AND the redirect in their `description`.** A\n   deprecated skill keeps a body that explicitly routes to the replacement —\n   never a placeholder — and the SAME routing text in its `description`, because\n   a loader REFUSES a SKILL.md whose description is absent or blank: the stub is\n   then never loaded, so its \"merged into X, load X instead\" body reaches nobody\n   and the tombstone is pure context cost with zero effect. An\n   `aliases: [\u003Cold-name>]` entry on the TARGET is the belt-and-braces form;\n   the stub alone is what actually resolves the old name. Deleting a redirect\n   without a forwarding target orphans every reference to the old name. The\n   doctor's `Skills loadable` check FAILs on exactly this class — a missing or\n   blank description, a name outside `[a-z0-9-]`, frontmatter that does not parse,\n   a duplicate mapping key PyYAML accepts but a loader rejects — so run it after\n   any consolidation pass instead of eyeballing the tombstones. **Check WHICH MODE that check ran\n   in — a degradation it does not announce is a PASS over files it never read.** The rules above\n   need a YAML parser, and the doctor runs under whatever `python3` is on PATH, which on a fleet\n   host is often a runtime venv without PyYAML; it then falls back to a TEXT scan and reports a\n   WARN naming the rules it skipped (name charset, YAML parse, duplicate keys). Read that warning\n   before trusting a bare `Skills loadable ✅`: it is a weaker result, not a clean one. Repair the\n   interpreter (`fix-fleet-venv.py`) and re-run, or invoke the doctor under `FLEET_PYTHON`\u002F\n   `~\u002F.steadfaste-fleet\u002Fvenv\u002Fbin\u002Fpython3`, so the full rule set is actually applied.\n   **But the durable fix lives in the CHECKER, not in the invocation.** Advice that says \"invoke this\n   under the right interpreter\" depends on every caller remembering, and the caller that forgets is\n   the one that gets a false green. A check should resolve its OWN interpreter — an explicit override\n   env var first, then the fleet venv, each probed by actually RUNNING it in a clean environment —\n   and re-exec itself once, guarded by an env flag so it cannot loop. A check that resolves for\n   itself cannot be run wrong.\n   The general rule for any check that depends on a library: a checker that CANNOT read its input\n   has no verdict — it must say so loudly, and it must never emit a success line for the inputs it\n   never examined, because a green tick over unread files is worse than a visible failure.\n4. **Check consumers before pruning.** A deployed-only skill may\n   be referenced by crons, manifests, or other skills. Grep jobs.json,\n   install-crons.sh, and skills before `delete`. Deploy warnings are not\n   deletes. A deployed-only skill is also an **orphan** — no repo source, so\n   the doctor reports `Stale skills: orphaned` and one deploy can remove it.\n   Import it into the repo or delete it deliberately; never leave it\n   half-owned, and same for a `references\u002F` file that exists only deployed.\n\n   **A third-party PLUGIN dump is a third case: neither import nor delete.** A\n   directory carrying its own plugin manifest (`package.json`,\n   `\u003Ctool>.plugin.json`, `_meta.json`) whose `SKILL.md` has no frontmatter has no\n   repo source to import and no owner left to repair it. Check the RUNTIME first\n   (`command -v \u003Ctool>`; its config directory) — the skills such a dump ships are\n   instructions to call tools that exist only while that plugin is loaded, so with\n   the runtime gone they are prompts to call nothing. Then PRESERVE it by MOVING it\n   into a dot-directory inside the same skills tree\n   (`\u003Cskills>\u002F.archive\u002F\u003Cname>-\u003Ctool>\u002F`) rather than deleting: loaders skip dot-dirs,\n   so it leaves every agent's view and stays recoverable. Assert\n   destination-absent before the move (never clobber), then assert the file COUNT\n   at the destination matches the source's.\n\n   **A ROOT-LEVEL skill directory is never pruned by a deploy** — the updater's\n   stale pass treats a path with no `\u002F` in it as a Hermes default and skips it. So\n   a removal like that is not undone by an update, and it does not propagate\n   either: another host carrying the same dump still gets the doctor WARN, which is\n   the correct behaviour — report it there, do not delete it fleet-wide from one\n   host's evidence.\n5. **Verify after every merge\u002Fdelete.** `skill_view` both skills, re-read\n   the merged SKILL.md for fence balance, run the doctor, regenerate the\n   manifest if skills\u002F changed.\n6. **The REPO is the source of truth — a curatorial edit made on the DEPLOYED\n   copy is STRANDED, not shipped.** `skill_manage` writes the DEPLOYED copy\n   (`~\u002F.hermes\u002Fskills\u002F\u003C...>`), while the fleet receives only\n   `steadfaste-fleet\u002Fskills\u002F\u003Ccategory>\u002F\u003Cname>\u002F`. A deploy then either overwrites\n   that copy or — where the drift guardrail is in force — SKIPS it; either way\n   the lesson reaches no other host, and the doctor reports it as skill drift\n   (*\"Deployed copy is newer than repo source\"*). The doctor names it: *\"Deployed copy is\n   newer than repo source. Commit the repo source before fleet-update\n   overwrites it.\"* After ANY skill edit, propagate the deployed skill back\n   to the repo path (the whole dir, `references\u002F` included) and commit it in\n   the same run. Check for drifted pairs at the **start** of a pass too: drift\n   means a previous run's lesson is still unlanded, and re-deriving it instead\n   of propagating it duplicates the work. Verify with `diff` — the only\n   acceptable difference is the deploy header.\n\n   **Measure drift by CONTENT, never by mtime.**\n   `python3 ops\u002Fscripts\u002Fmanage\u002Fcheck-skill-drift-parity.py` (exit 1 = stranded;\n   regenerates `docs\u002Fevidence\u002Fskill-drift-parity.txt`). A file is stranded when its\n   DEPLOYED content appears neither in the repo working tree nor in any committed\n   revision of that path. mtime is useless here — a clone or checkout resets every repo\n   mtime to \"now\", so an mtime comparison reports \"repo is newer\" while the content\n   exists nowhere in the repo and the drift goes unnoticed. It compares EVERY file under\n   the deployed tree, not just `SKILL.md`, so a drifted `references\u002F` file is visible too.\n\n   **Decide the direction PER FILE by reading the diff — the deployed side is often the\n   CURATED one.** A deployed lesson may have been rewritten to generalise, so it can ADD\n   lines AND REMOVE lines the repo still has: never copy it in blind, and never assume\n   the repo side is the superset. The deploy header sits AFTER the shebang and is closed\n   by a blank line, so a fixed `tail -n +4` misaligns the comparison — strip the header\n   LINES and diff what remains. Copy the canonical content in, re-run the parity check to\n   `PASS`, then deploy to clear the doctor's warnings.\n\n   **Recover with the repo's own tooling — and judge its output by CONTENT, not by its\n   own line counts.** Run `python3 ops\u002Fscripts\u002Fmanage\u002Freconcile-stranded-skills.py`\n   (report by default) and land the pure-superset class with `--apply`. Its REVIEW class\n   is not a refusal to act: the tool compares whole LINES, so a paragraph that was\n   REWRAPPED or condensed reports every one of its lines as a \"drop\". Collapse\n   whitespace on both sides and word-diff instead — most REVIEW entries then show ZERO\n   repo-only word runs and are pure supersets in substance. Make the copy DECIDABLE before making\n   it: run both sides through `difflib.SequenceMatcher` and collect the non-`equal` opcodes — at\n   most ONE line-dropping block (an in-place rewrite) plus insertions means the deployed side is\n   that content, so copy it; two or more dropping blocks means the newer copy genuinely REMOVED\n   content you must read first. Assert the count rather than eyeballing a diffstat, and keep the\n   assertion in the script so a later reconcile cannot quietly take a side it should have read.\n   Only a phrase genuinely\n   absent from the deployed copy needs a decision, and the answer is often \"superseded in\n   place\": a newer copy that CORRECTS a wrong claim (an obsolete design description, a\n   rule reworded to include the mechanism the fleet now uses) must be landed AS-IS,\n   because merging the old wording back restores the error it removed. Land by copying\n   the file and ASSERTING sha256 equality on both sides, so byte-identical is verified\n   rather than claimed. Re-run the report afterwards: it must say `stranded: 0`.\n\n   **Two classes, and the doctor only reports one of them.** \"Stranded in place\" (the\n   path IS in the repo, its content only deployed) is what the doctor's `Skill drift`\n   line counts, and the line flips to PASS when that reaches 0. The parity checker ALSO\n   reports \"absent from the repo\" — files under a repo-owned skill dir the repo has never\n   carried at all, marked `dropped-from-repo` or `never-committed`. Driving the first to\n   0 leaves the checker's VERDICT at FAIL, so read both numbers before reporting the\n   class as fixed, and keep the tracked `docs\u002Fevidence\u002Fskill-drift-parity.txt`\n   regenerated and committed (it records the verdict; leaving it stale is the same defect\n   one level up).\n\n   **Recovering a skill's SCRIPT is the step that stalls: the pre-commit TDD gate refuses\n   an ADDED code file with no test in the same commit.** The gate scans\n   `--diff-filter=A` for `py|ts|tsx|js|jsx|mjs|cjs|go|rs|…` and has no bypass flag, and a\n   `skills\u002F\u003Cname>\u002Fscripts\u002F\u003Cfile>.py` is exactly that — so the asset sits uncommitted\n   indefinitely, which is a large part of why this class accumulates. Do NOT stage an\n   unrelated test to satisfy it; add the guard the gate is actually asking for.\n   `tests\u002Ftest_skill_code_assets_parse.py` is that guard: one discovery traversal over\n   `skills\u002F**` that compiles `.py` and parse-checks `.mjs`\u002F`.js`\u002F`.sh`, so a newly\n   recovered asset is covered the moment it lands. Extend it rather than writing a second\n   one, and keep its non-vacuity assertion (it FAILs when it finds no assets at all).\n\n   **Read a recovered DOCUMENT before landing it — the recovery leaves defects of its own.** A\n   recovered reference is not good because it came from a host where it worked: a redaction pass can\n   destroy the very distinction the document teaches (a fix example whose \"before\" and \"after\" lines\n   are byte-identical because both were rewritten to the placeholder), and a recovered script can\n   carry the bug it warns about (reading a column by hardcoded INDEX while `PRAGMA table_info` was\n   already fetched). These are cheap to fix while landing and expensive later: the close judges each\n   recovered file as its own blob, so a self-contradicting document is found one blob and one review\n   round at a time, and every fix commit grows a range that was already too wide. Read for the\n   artifact of the RECOVERY, not only for the lesson.\n\n   **Price a bulk recovery by CLAIM, not by file count.** The close reviewer judges what each\n   recovered blob CLAIMS. A document that presents an incident note as a live statement about code\n   is judged as BEHAVIOUR, so a committed re-runnable test is demanded — which cannot exist for a\n   past measurement, and the file's real claim is a POINTER to a record instead. Land recovered\n   history declaring itself as history (what it is a record OF, and as-of when) rather than as\n   specification, and land recovered prose in WAVES: a range whose files each need a verdict costs\n   passes in proportion to its FILE count, so a one-hundred-file recovery commit is the most\n   expensive shape available. Expect the class and diagnose it as a claim-type mismatch, not as a\n   fault in the recovered text.\n\n7. **A same-name collision from a `local-*` skill is the OVERRIDE WORKING — do not\n   \"fix\" it.** A host-local `local-\u003Cx>` skill deliberately keeps the fleet skill's\n   `name: \u003Cx>`, so it wins the name collision and the fleet copy is skipped: that IS\n   the per-host override mechanism, and the harness reports it as a collision\n   warning rather than an error. Renaming the local skill's `name:` to match its\n   directory makes BOTH load and the override silently stops working.\n   `fleet-update.sh`'s stale pass preserves every `local-*` directory for the same\n   reason. A `local-*` skill that carries a real frontmatter defect (a duplicate\n   key, a name outside the charset) is still broken, though — the collision is\n   expected, the parse failure is not.\n\n   **How a collision arises at all — the name falls back to the DIRECTORY.** A\n   loader uses `frontmatter.name`, and when that is absent it uses the skill\n   directory's own basename, so a directory carrying no `name:` is advertised under\n   its DIRECTORY name and can collide with a different skill that declares that name\n   in frontmatter. On collision the loser is skipped and the winner is whichever the\n   traversal reached first — decided by readdir order, not by intent — which is\n   precisely the mechanism a `local-*` override exploits on purpose. Note the\n   interaction with rule 3: a SKILL.md with no DESCRIPTION never reaches the name\n   map at all, so a no-frontmatter dump is dropped outright and its only effects are\n   the diagnostic and the index cost. Read the loader's own DIAGNOSTICS LIST before\n   calling something a collision — the two failures look identical in a directory\n   listing and are different findings.\n\n   **A stray non-skill `.md` in the tree is a PHANTOM candidate.** A `DESCRIPTION.md`, `README.md`\n   or similar sitting inside a category directory is enumerated as a skill whose name falls back to\n   the DIRECTORY, so `skills\u002F\u003Ccat>\u002FDESCRIPTION.md` presents as a second skill named after its\n   category and collides with the real one — and since the winner is readdir order, the phantom can\n   win. Keep non-skill markdown out of the skill tree, and read a collision whose loser is a\n   frontmatter-less `.md` as this shape rather than as a deliberate override.\n\n8. **A retired skill name is not done at the stub — sweep every REFERENCE to it, then guard it.**\n   A stub only helps a caller who already holds the old name; a reference in another skill's\n   prose or `related_skills` costs a wasted load BEFORE the redirect fires. A consolidation\n   pass therefore ends with a sweep and a check, not with the tombstone.\n   - **Sweep the two places a `SKILL.md`-only scan cannot see:** backticked names in\n     `references\u002F` and `templates\u002F`, and `related_skills` read as PARSED YAML — the list may\n     be nested under `metadata.hermes` or span several lines, so a single inline\n     `related_skills: [...]` grep pattern finds barely half the hits. Parse the frontmatter\n     and collect from BOTH locations.\n   - **Rewrite each hit to the stub's `absorbed_into`\u002F`replaced_by` target, and require that\n     field on every stub.** With it the sweep is mechanical and assertable; without it every\n     hit is a judgement call about where the content actually went. Assert the replacement\n     COUNT per file, then re-scan for zero remaining.\n   - **Scope the guard to a CLOSED set.** Match only names that actually carry\n     `deprecated: true`, so the check cannot fire on prose that merely looks name-like. The\n     tempting version — \"every `related_skills` entry must resolve inside this tree\" — is\n     noisy: a skill can be provided from a tree you are not scanning, and some entries name a\n     TOOL rather than a skill. Precision is the feature; a guard that cries wolf teaches its\n     reader to scroll past it.\n   - **Check repeats PER LIST, not across merged lists.** A file may legitimately declare the\n     same entries at the top level AND under `metadata.hermes`; merging them reports a phantom\n     duplicate. The defect worth checking is a repeat WITHIN one list.\n   - **Scan for what the sweep LEFT BEHIND, not only for what it matched.** Redirecting\n     `old` -> `new` creates a duplicate list entry whenever the list already named `new`, and\n     collapses two related bullets onto one skill when both pointed at names redirecting to\n     the same target; a gloss describing the OLD skill then sits under the new name and must\n     be rewritten to match — the redirect moves the annotation too. An asserted match count\n     catches none of these, because they are consequences of a CORRECT replacement, so the\n     sweep needs its own second pass looking for them.\n\n## Manifest Discipline\n\n- After any skill add\u002Frename\u002Fdelete: regenerate\n  `docs\u002FSKILLS-MANIFEST.md` via `gen-skills-manifest.py` (never hand-edit).\n- Keep category counts truthful — a stub or placeholder body in the repo is\n  a curation debt: flag for restore, don't let it ship.\n\n## Anti-Patterns\n\n- Merging by description without reading both files\n- Deleting a skill and leaving its references dangling\n- \"Fixing\" a balanced file's fences (see orch-skill-lifecycle fence rule)\n- Creating a new skill because the existing one needed a patch\n\n## Bulk Curation Mechanics\n\n- `skill_manage` **`delete` is a sole op** — it refuses to compose with other\n  ops in a single `operations` array (rollback safety). Fire each `delete` as\n  its own `skill_manage` call; issue several single-op calls in parallel when\n  pruning a batch, but never put a `delete` alongside another action.\n- Writing redirect bodies over many existing `SKILL.md` files hits the\n  `write_file` read-before-write guard: it refuses to overwrite a file whose\n  current content you have not fully read this task, and a `read_file(limit=N)`\n  counts as a \"partial view\" that still blocks. For bulk redirects, delegate\n  the mechanical overwrite to a subagent (it reads-then-writes cleanly), or\n  read each file fully (no `limit`) first, or `patch` just the frontmatter\n  lines instead of overwriting the whole file.\n","# Skill Curation\n\nConsolidate, dedupe, and prune the skill library. Curation is the\ncounterweight to creation: every new skill is debt until it earns its place,\nand a merged skill serves better than two overlapping ones.\n\n## When to Curate\n\n- After a lifecycle run (orch-skill-lifecycle Monday deep eval)\n- When 2+ skills overlap — `skills_list(category=...)` shows same-domain\n  entries with similar descriptions\n- On any deletion — always declare `absorbed_into=\u003Cumbrella>` (merged\n  into an existing skill) or `absorbed_into=\"\"` (true prune) so downstream\n  tooling (cron-job skill references, the curator) knows intent\n- When the doctor flags skill drift, duplicates, or deployed orphans\n\n## Rules\n\n1. **Prove duplication first.** Load both skills and compare structure AND\n   content — descriptions alone lie. Overlap > 50% → merge; otherwise keep\n   both and note the boundary.\n2. **Improve existing before creating.** The merge target is the existing\n   skill — extend it, never spawn a sibling. If the two are equal weight,\n   keep the one with better structure, fold the other's unique sections in.\n3. **Redirects carry real bodies AND the redirect in their `description`.** A\n   deprecated skill keeps a body that explicitly routes to the replacement —\n   never a placeholder — and the SAME routing text in its `description`, because\n   a loader REFUSES a SKILL.md whose description is absent or blank: the stub is\n   then never loaded, so its \"merged into X, load X instead\" body reaches nobody\n   and the tombstone is pure context cost with zero effect. An\n   `aliases: [\u003Cold-name>]` entry on the TARGET is the belt-and-braces form;\n   the stub alone is what actually resolves the old name. Deleting a redirect\n   without a forwarding target orphans every reference to the old name. The\n   doctor's `Skills loadable` check FAILs on exactly this class — a missing or\n   blank description, a name outside `[a-z0-9-]`, frontmatter that does not parse,\n   a duplicate mapping key PyYAML accepts but a loader rejects — so run it after\n   any consolidation pass instead of eyeballing the tombstones. **Check WHICH MODE that check ran\n   in — a degradation it does not announce is a PASS over files it never read.** The rules above\n   need a YAML parser, and the doctor runs under whatever `python3` is on PATH, which on a fleet\n   host is often a runtime venv without PyYAML; it then falls back to a TEXT scan and reports a\n   WARN naming the rules it skipped (name charset, YAML parse, duplicate keys). Read that warning\n   before trusting a bare `Skills loadable ✅`: it is a weaker result, not a clean one. Repair the\n   interpreter (`fix-fleet-venv.py`) and re-run, or invoke the doctor under `FLEET_PYTHON`\u002F\n   `~\u002F.steadfaste-fleet\u002Fvenv\u002Fbin\u002Fpython3`, so the full rule set is actually applied.\n   **But the durable fix lives in the CHECKER, not in the invocation.** Advice that says \"invoke this\n   under the right interpreter\" depends on every caller remembering, and the caller that forgets is\n   the one that gets a false green. A check should resolve its OWN interpreter — an explicit override\n   env var first, then the fleet venv, each probed by actually RUNNING it in a clean environment —\n   and re-exec itself once, guarded by an env flag so it cannot loop. A check that resolves for\n   itself cannot be run wrong.\n   The general rule for any check that depends on a library: a checker that CANNOT read its input\n   has no verdict — it must say so loudly, and it must never emit a success line for the inputs it\n   never examined, because a green tick over unread files is worse than a visible failure.\n4. **Check consumers before pruning.** A deployed-only skill may\n   be referenced by crons, manifests, or other skills. Grep jobs.json,\n   install-crons.sh, and skills before `delete`. Deploy warnings are not\n   deletes. A deployed-only skill is also an **orphan** — no repo source, so\n   the doctor reports `Stale skills: orphaned` and one deploy can remove it.\n   Import it into the repo or delete it deliberately; never leave it\n   half-owned, and same for a `references\u002F` file that exists only deployed.\n\n   **A third-party PLUGIN dump is a third case: neither import nor delete.** A\n   directory carrying its own plugin manifest (`package.json`,\n   `\u003Ctool>.plugin.json`, `_meta.json`) whose `SKILL.md` has no frontmatter has no\n   repo source to import and no owner left to repair it. Check the RUNTIME first\n   (`command -v \u003Ctool>`; its config directory) — the skills such a dump ships are\n   instructions to call tools that exist only while that plugin is loaded, so with\n   the runtime gone they are prompts to call nothing. Then PRESERVE it by MOVING it\n   into a dot-directory inside the same skills tree\n   (`\u003Cskills>\u002F.archive\u002F\u003Cname>-\u003Ctool>\u002F`) rather than deleting: loaders skip dot-dirs,\n   so it leaves every agent's view and stays recoverable. Assert\n   destination-absent before the move (never clobber), then assert the file COUNT\n   at the destination matches the source's.\n\n   **A ROOT-LEVEL skill directory is never pruned by a deploy** — the updater's\n   stale pass treats a path with no `\u002F` in it as a Hermes default and skips it. So\n   a removal like that is not undone by an update, and it does not propagate\n   either: another host carrying the same dump still gets the doctor WARN, which is\n   the correct behaviour — report it there, do not delete it fleet-wide from one\n   host's evidence.\n5. **Verify after every merge\u002Fdelete.** `skill_view` both skills, re-read\n   the merged SKILL.md for fence balance, run the doctor, regenerate the\n   manifest if skills\u002F changed.\n6. **The REPO is the source of truth — a curatorial edit made on the DEPLOYED\n   copy is STRANDED, not shipped.** `skill_manage` writes the DEPLOYED copy\n   (`~\u002F.hermes\u002Fskills\u002F\u003C...>`), while the fleet receives only\n   `steadfaste-fleet\u002Fskills\u002F\u003Ccategory>\u002F\u003Cname>\u002F`. A deploy then either overwrites\n   that copy or — where the drift guardrail is in force — SKIPS it; either way\n   the lesson reaches no other host, and the doctor reports it as skill drift\n   (*\"Deployed copy is newer than repo source\"*). The doctor names it: *\"Deployed copy is\n   newer than repo source. Commit the repo source before fleet-update\n   overwrites it.\"* After ANY skill edit, propagate the deployed skill back\n   to the repo path (the whole dir, `references\u002F` included) and commit it in\n   the same run. Check for drifted pairs at the **start** of a pass too: drift\n   means a previous run's lesson is still unlanded, and re-deriving it instead\n   of propagating it duplicates the work. Verify with `diff` — the only\n   acceptable difference is the deploy header.\n\n   **Measure drift by CONTENT, never by mtime.**\n   `python3 ops\u002Fscripts\u002Fmanage\u002Fcheck-skill-drift-parity.py` (exit 1 = stranded;\n   regenerates `docs\u002Fevidence\u002Fskill-drift-parity.txt`). A file is stranded when its\n   DEPLOYED content appears neither in the repo working tree nor in any committed\n   revision of that path. mtime is useless here — a clone or checkout resets every repo\n   mtime to \"now\", so an mtime comparison reports \"repo is newer\" while the content\n   exists nowhere in the repo and the drift goes unnoticed. It compares EVERY file under\n   the deployed tree, not just `SKILL.md`, so a drifted `references\u002F` file is visible too.\n\n   **Decide the direction PER FILE by reading the diff — the deployed side is often the\n   CURATED one.** A deployed lesson may have been rewritten to generalise, so it can ADD\n   lines AND REMOVE lines the repo still has: never copy it in blind, and never assume\n   the repo side is the superset. The deploy header sits AFTER the shebang and is closed\n   by a blank line, so a fixed `tail -n +4` misaligns the comparison — strip the header\n   LINES and diff what remains. Copy the canonical content in, re-run the parity check to\n   `PASS`, then deploy to clear the doctor's warnings.\n\n   **Recover with the repo's own tooling — and judge its output by CONTENT, not by its\n   own line counts.** Run `python3 ops\u002Fscripts\u002Fmanage\u002Freconcile-stranded-skills.py`\n   (report by default) and land the pure-superset class with `--apply`. Its REVIEW class\n   is not a refusal to act: the tool compares whole LINES, so a paragraph that was\n   REWRAPPED or condensed reports every one of its lines as a \"drop\". Collapse\n   whitespace on both sides and word-diff instead — most REVIEW entries then show ZERO\n   repo-only word runs and are pure supersets in substance. Make the copy DECIDABLE before making\n   it: run both sides through `difflib.SequenceMatcher` and collect the non-`equal` opcodes — at\n   most ONE line-dropping block (an in-place rewrite) plus insertions means the deployed side is\n   that content, so copy it; two or more dropping blocks means the newer copy genuinely REMOVED\n   content you must read first. Assert the count rather than eyeballing a diffstat, and keep the\n   assertion in the script so a later reconcile cannot quietly take a side it should have read.\n   Only a phrase genuinely\n   absent from the deployed copy needs a decision, and the answer is often \"superseded in\n   place\": a newer copy that CORRECTS a wrong claim (an obsolete design description, a\n   rule reworded to include the mechanism the fleet now uses) must be landed AS-IS,\n   because merging the old wording back restores the error it removed. Land by copying\n   the file and ASSERTING sha256 equality on both sides, so byte-identical is verified\n   rather than claimed. Re-run the report afterwards: it must say `stranded: 0`.\n\n   **Two classes, and the doctor only reports one of them.** \"Stranded in place\" (the\n   path IS in the repo, its content only deployed) is what the doctor's `Skill drift`\n   line counts, and the line flips to PASS when that reaches 0. The parity checker ALSO\n   reports \"absent from the repo\" — files under a repo-owned skill dir the repo has never\n   carried at all, marked `dropped-from-repo` or `never-committed`. Driving the first to\n   0 leaves the checker's VERDICT at FAIL, so read both numbers before reporting the\n   class as fixed, and keep the tracked `docs\u002Fevidence\u002Fskill-drift-parity.txt`\n   regenerated and committed (it records the verdict; leaving it stale is the same defect\n   one level up).\n\n   **Recovering a skill's SCRIPT is the step that stalls: the pre-commit TDD gate refuses\n   an ADDED code file with no test in the same commit.** The gate scans\n   `--diff-filter=A` for `py|ts|tsx|js|jsx|mjs|cjs|go|rs|…` and has no bypass flag, and a\n   `skills\u002F\u003Cname>\u002Fscripts\u002F\u003Cfile>.py` is exactly that — so the asset sits uncommitted\n   indefinitely, which is a large part of why this class accumulates. Do NOT stage an\n   unrelated test to satisfy it; add the guard the gate is actually asking for.\n   `tests\u002Ftest_skill_code_assets_parse.py` is that guard: one discovery traversal over\n   `skills\u002F**` that compiles `.py` and parse-checks `.mjs`\u002F`.js`\u002F`.sh`, so a newly\n   recovered asset is covered the moment it lands. Extend it rather than writing a second\n   one, and keep its non-vacuity assertion (it FAILs when it finds no assets at all).\n\n   **Read a recovered DOCUMENT before landing it — the recovery leaves defects of its own.** A\n   recovered reference is not good because it came from a host where it worked: a redaction pass can\n   destroy the very distinction the document teaches (a fix example whose \"before\" and \"after\" lines\n   are byte-identical because both were rewritten to the placeholder), and a recovered script can\n   carry the bug it warns about (reading a column by hardcoded INDEX while `PRAGMA table_info` was\n   already fetched). These are cheap to fix while landing and expensive later: the close judges each\n   recovered file as its own blob, so a self-contradicting document is found one blob and one review\n   round at a time, and every fix commit grows a range that was already too wide. Read for the\n   artifact of the RECOVERY, not only for the lesson.\n\n   **Price a bulk recovery by CLAIM, not by file count.** The close reviewer judges what each\n   recovered blob CLAIMS. A document that presents an incident note as a live statement about code\n   is judged as BEHAVIOUR, so a committed re-runnable test is demanded — which cannot exist for a\n   past measurement, and the file's real claim is a POINTER to a record instead. Land recovered\n   history declaring itself as history (what it is a record OF, and as-of when) rather than as\n   specification, and land recovered prose in WAVES: a range whose files each need a verdict costs\n   passes in proportion to its FILE count, so a one-hundred-file recovery commit is the most\n   expensive shape available. Expect the class and diagnose it as a claim-type mismatch, not as a\n   fault in the recovered text.\n\n7. **A same-name collision from a `local-*` skill is the OVERRIDE WORKING — do not\n   \"fix\" it.** A host-local `local-\u003Cx>` skill deliberately keeps the fleet skill's\n   `name: \u003Cx>`, so it wins the name collision and the fleet copy is skipped: that IS\n   the per-host override mechanism, and the harness reports it as a collision\n   warning rather than an error. Renaming the local skill's `name:` to match its\n   directory makes BOTH load and the override silently stops working.\n   `fleet-update.sh`'s stale pass preserves every `local-*` directory for the same\n   reason. A `local-*` skill that carries a real frontmatter defect (a duplicate\n   key, a name outside the charset) is still broken, though — the collision is\n   expected, the parse failure is not.\n\n   **How a collision arises at all — the name falls back to the DIRECTORY.** A\n   loader uses `frontmatter.name`, and when that is absent it uses the skill\n   directory's own basename, so a directory carrying no `name:` is advertised under\n   its DIRECTORY name and can collide with a different skill that declares that name\n   in frontmatter. On collision the loser is skipped and the winner is whichever the\n   traversal reached first — decided by readdir order, not by intent — which is\n   precisely the mechanism a `local-*` override exploits on purpose. Note the\n   interaction with rule 3: a SKILL.md with no DESCRIPTION never reaches the name\n   map at all, so a no-frontmatter dump is dropped outright and its only effects are\n   the diagnostic and the index cost. Read the loader's own DIAGNOSTICS LIST before\n   calling something a collision — the two failures look identical in a directory\n   listing and are different findings.\n\n   **A stray non-skill `.md` in the tree is a PHANTOM candidate.** A `DESCRIPTION.md`, `README.md`\n   or similar sitting inside a category directory is enumerated as a skill whose name falls back to\n   the DIRECTORY, so `skills\u002F\u003Ccat>\u002FDESCRIPTION.md` presents as a second skill named after its\n   category and collides with the real one — and since the winner is readdir order, the phantom can\n   win. Keep non-skill markdown out of the skill tree, and read a collision whose loser is a\n   frontmatter-less `.md` as this shape rather than as a deliberate override.\n\n8. **A retired skill name is not done at the stub — sweep every REFERENCE to it, then guard it.**\n   A stub only helps a caller who already holds the old name; a reference in another skill's\n   prose or `related_skills` costs a wasted load BEFORE the redirect fires. A consolidation\n   pass therefore ends with a sweep and a check, not with the tombstone.\n   - **Sweep the two places a `SKILL.md`-only scan cannot see:** backticked names in\n     `references\u002F` and `templates\u002F`, and `related_skills` read as PARSED YAML — the list may\n     be nested under `metadata.hermes` or span several lines, so a single inline\n     `related_skills: [...]` grep pattern finds barely half the hits. Parse the frontmatter\n     and collect from BOTH locations.\n   - **Rewrite each hit to the stub's `absorbed_into`\u002F`replaced_by` target, and require that\n     field on every stub.** With it the sweep is mechanical and assertable; without it every\n     hit is a judgement call about where the content actually went. Assert the replacement\n     COUNT per file, then re-scan for zero remaining.\n   - **Scope the guard to a CLOSED set.** Match only names that actually carry\n     `deprecated: true`, so the check cannot fire on prose that merely looks name-like. The\n     tempting version — \"every `related_skills` entry must resolve inside this tree\" — is\n     noisy: a skill can be provided from a tree you are not scanning, and some entries name a\n     TOOL rather than a skill. Precision is the feature; a guard that cries wolf teaches its\n     reader to scroll past it.\n   - **Check repeats PER LIST, not across merged lists.** A file may legitimately declare the\n     same entries at the top level AND under `metadata.hermes`; merging them reports a phantom\n     duplicate. The defect worth checking is a repeat WITHIN one list.\n   - **Scan for what the sweep LEFT BEHIND, not only for what it matched.** Redirecting\n     `old` -> `new` creates a duplicate list entry whenever the list already named `new`, and\n     collapses two related bullets onto one skill when both pointed at names redirecting to\n     the same target; a gloss describing the OLD skill then sits under the new name and must\n     be rewritten to match — the redirect moves the annotation too. An asserted match count\n     catches none of these, because they are consequences of a CORRECT replacement, so the\n     sweep needs its own second pass looking for them.\n\n## Manifest Discipline\n\n- After any skill add\u002Frename\u002Fdelete: regenerate\n  `docs\u002FSKILLS-MANIFEST.md` via `gen-skills-manifest.py` (never hand-edit).\n- Keep category counts truthful — a stub or placeholder body in the repo is\n  a curation debt: flag for restore, don't let it ship.\n\n## Anti-Patterns\n\n- Merging by description without reading both files\n- Deleting a skill and leaving its references dangling\n- \"Fixing\" a balanced file's fences (see orch-skill-lifecycle fence rule)\n- Creating a new skill because the existing one needed a patch\n\n## Bulk Curation Mechanics\n\n- `skill_manage` **`delete` is a sole op** — it refuses to compose with other\n  ops in a single `operations` array (rollback safety). Fire each `delete` as\n  its own `skill_manage` call; issue several single-op calls in parallel when\n  pruning a batch, but never put a `delete` alongside another action.\n- Writing redirect bodies over many existing `SKILL.md` files hits the\n  `write_file` read-before-write guard: it refuses to overwrite a file whose\n  current content you have not fully read this task, and a `read_file(limit=N)`\n  counts as a \"partial view\" that still blocks. For bulk redirects, delegate\n  the mechanical overwrite to a subagent (it reads-then-writes cleanly), or\n  read each file fully (no `limit`) first, or `patch` just the frontmatter\n  lines instead of overwriting the whole file.\n",{"name":5,"version":24,"triggers":25,"description":7},"1.1.0",[26,27,28,29,30],"curate skills","consolidate skills","merge skills","prune skills","skill dedupe","main","9dc0e92c5fa81d4db5292b205716e8ea0c00d10e","skills\u002Fdevops\u002Fskill-curation",1,"https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet","steadfaste.com","https:\u002F\u002Fgithub.com\u002Flukemcqueen","AI Architect | Building Hermes Cortex: governed multi-agent AI for autonomous software engineering | Enterprise Tech Consultant","2026-11-14 22:33:34",{"source":8,"skillName":5,"commands":41,"prompt":51,"zipAvailable":52},[42,45,48],{"tool":43,"command":44},"npx","npx skills add lukemcqueen\u002Fsteadfaste-fleet --skill skill-curation",{"tool":46,"command":47},"bunx","bunx skills add lukemcqueen\u002Fsteadfaste-fleet --skill skill-curation",{"tool":49,"command":50},"pnpm","pnpm dlx skills add lukemcqueen\u002Fsteadfaste-fleet --skill skill-curation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fskill-curation\u002FSKILL.md",true,[54,60,64],{"path":55,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Falways-section-consolidation.md","REFERENCE","参考文档",4690,"https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fskill-curation\u002Freferences\u002Falways-section-consolidation.md",{"path":61,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":62,"sourceHtmlUrl":63},"references\u002Fdiscovery-sources.md",2794,"https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fskill-curation\u002Freferences\u002Fdiscovery-sources.md",{"path":65,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":66,"sourceHtmlUrl":67},"references\u002Fskill-evaluation-pipeline.md",8914,"https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fskill-curation\u002Freferences\u002Fskill-evaluation-pipeline.md",[69,77,84,91,98,105],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":76},"01M7DXY4G8NSVB59K8B6S2AQNP","macbook-linux-maintenance","Legacy MacBook Linux fixes: GPU wedge, reboot hang, thermal.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fmacbook-linux-maintenance\u002FSKILL.md",[],[],"2026-11-15 14:19:17",{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":76},"01M7DXXYJCXFJ3TGB84ZQ8KFE6","governance-regression-triage","Triage a host's governance issues: what changed.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fgovernance-regression-triage\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":76},"01M7DXXSAP1G8FDRJAMXQGBBMK","enforcement-immutability","Use when enforcement files lose their immutable flag.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fenforcement-immutability\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":76},"01M78SSDQVRYX64Y16X32TC0NS","steadfaste-fleet-repo-deploy","Use when committing to the steadfaste-fleet repo.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fsteadfaste-fleet-repo-deploy\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":76},"01M78SS928F3R2RJAYVW4E30CE","steadfaste-fleet-maintenance","Maintain an installed Steadfaste Fleet instance — update both the upstream Hermes Agent and the fleet repo layer, merge diverged histories, sync skills safely, and troubleshoot common issues. Covers the dual-update workflow (hermes update + fleet-update.sh), encrypted brain data backup, mycortex import, and recovery techniques.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fsteadfaste-fleet-maintenance\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"nameZh":6,"description":108,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":109,"categories":110,"keywords":111,"publishedAt":76},"01M78SP1CHW0T0BWE5ZT6JBPX8","fleet-preflight","DEPRECATED alias — merged into survey-before-action (2026-08-20). Load survey-before-action instead; its \"Repo-Specific Pre-Flight\" section covers git search, Hermes boundary, and deployment verification.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Ffleet-preflight\u002FSKILL.md",[],[]]