[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7AX9S9V95QPP5QE2SW5HDPS":3,"skill-same:01M7AX9S9V95QPP5QE2SW5HDPS":46},{"skillNo":4,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":29,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M7AX9S9V95QPP5QE2SW5HDPS","harness-integration",null,"roryeckel\u002Fhooknostic","roryeckel","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F16675082?v=4","01M7AX8JWEFHND7P2G338X0TQW","Rory","01M7AX8JX2FGW3TMBVSGFAPQD2",2,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fharness-integration\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: harness-integration\ndescription: Plan and land complete support for a new coding-agent harness, or a new delivery surface for one. Use when adding or auditing support for Cursor, Gemini CLI, Goose, GitHub Copilot CLI, or any other agent CLI; when creating `packages\u002Fadapter-\u003Cid>\u002F`, `fixtures\u002F\u003Charness>\u002F`, `.capture\u002F\u003Charness>\u002F`, an adapter profile, a shim, an Agent Plugin projector, a project integrator, or a harness-watch lane; or when an adapter must grow hooks, skills, MCP, package\u002Fproject delivery, playback, drift, or generated docs. Also use when a harness reads another harness's files (a compatibility mode: Copilot loading `.claude\u002Fsettings.json` hooks, `.claude\u002Fskills` or a root `.mcp.json`), when one target's output runs under another harness, or when changing where a project integration writes. Start with `docs\u002Fadding-an-adapter.md` and the `harness-capture` skill; this skill is the planning, coverage, and completeness layer.\n---\n\n# Harness integration\n\nSupport is a matrix of evidence-backed answers, not a compiling adapter. A\nharness is integrated when every surface it exposes is either implemented and\nproven, or explicitly declined with a reason and a provenance record. \"It\nbuilds\" and \"the docs say so\" are not completion criteria.\n\nThis skill is the orchestration and completeness layer. The implementation\nmechanics live in `docs\u002Fadding-an-adapter.md`; capture technique lives in\n`.agents\u002Fskills\u002Fharness-capture\u002FSKILL.md`; vocabulary changes live in\n`.agents\u002Fskills\u002Fextend-vocabulary\u002FSKILL.md`.\n\n## Required context\n\nRead before planning:\n\n- `AGENTS.md` and `CONTRIBUTING.md` — hard rules, scope, verification discipline.\n- `docs\u002Fadding-an-adapter.md` — the authoritative adapter workflow.\n- `docs\u002Fdesign.md` Appendix C — the definition of done.\n- `docs\u002Ftesting.md` — fixture replay, real-harness playback, pty and smoke layers.\n- `docs\u002Fharness-watch.md` — scheduled verification, drift, record, merge lanes.\n- Relevant ADRs in `docs\u002Fdecisions\u002F`: 0007 (tool shapes), 0008 (version facts),\n  0009 (rolling playback record), 0010 (scenario coverage), 0011 (Agent Plugin\n  projection), 0019 (deviations), 0020 (plugin root), 0021 (degradations),\n  0022 (shortfall policy), 0026 (file shapes), 0027 (optional fields),\n  0028 (skill directory), 0029 (shell dialect), 0030 (skill invocation policy),\n  0031 (named omissions), 0032 (MCP identity).\n\n## Phase 0 — Plan the integration\n\nBuild a surface matrix before writing code. The plan is a list of questions and\nthe evidence that will answer each one, not a list of files.\n\n### Choose the reference truth\n\n- Pick the harness id, binary or npm package, reference build, and fixture\n  directory name. Record them in `harness.ts`; every test and range derives from\n  that metadata.\n- If an older release is known broken, set the recommended range to the proven\n  boundary and propagate it to capability and projector profiles. Narrowing one\n  component while another still claims the broken range ships silent failures.\n- Record the platform matrix. A Windows-only capture does not establish Linux\n  shell behavior, and a headless capture does not establish an interactive\n  path. Plan a container or a second host where the harness ships one.\n- Decide the delivery modes: package, project, or both. A harness can have a\n  working package route and no project route, or the reverse. Do not assume one\n  from the other.\n- Decide whether the harness has a plugin\u002FAgent Plugins format, an in-process\n  extension model, or only command hooks. This determines the shim shape.\n\n### Surface inventory\n\nCopy this into the plan and fill every row with `capture`, `declined`, or `not\napplicable`. A blank row is an unowned risk.\n\n| Surface | Questions to answer | Evidence |\n| --- | --- | --- |\n| Process model | Command hook, in-process callback, imported module, long-lived server? What invokes it; what is stdin\u002Fstdout\u002Fargv; what is the timeout and failure mode? | Live harness session |\n| Hook envelope | Event names in every spelling; session\u002Fcwd\u002Ftimestamp\u002Fcorrelation fields; does the payload name its event? | Captured payload |\n| Lifecycle | Session start\u002Fend, resume, source, model, cwd, worktree, trust, folder-open behavior | Passive capture |\n| Prompt | User prompt before\u002Fafter transformation, output channels, block\u002Fcontext delivery | Passive + effect capture |\n| Tool before\u002Fafter | Native names, argument keys, result shapes, errors, parallel\u002Fbackground calls, MCP names | Tool-shape capture |\n| Effects | Block\u002Fdeny, ask\u002Fapproval, input rewrite, output rewrite, context injection, stop prevention, notification, session context | One effect-proof per channel |\n| Shell | Every shell tool advertised on each host; command\u002Fcwd keys; interpreter identity and dialect | Payload + process identity |\n| File tools | Read\u002Fwrite\u002Fedit\u002Fpatch\u002Fsearch shapes; path key or patch grammar; read-side limits | Captured tool calls |\n| Skills | Project and personal discovery; package discovery; body\u002Fpath expansion; frontmatter; explicit vs implicit invocation; native gate true\u002Ffalse; plugin namespace; in-place sources | Interactive TUI + `-p` controls |\n| MCP | stdio, Streamable HTTP, SSE; command\u002Fargs\u002Fenv\u002Fcwd; remote headers; placeholder expansion; startup; tool identity; runtime dependencies | Project and package probes |\n| Agent Plugins\u002Fplugin | Manifest, namespace, hooks, commands, agents, rules, LSP, client extensions, runtime package, load\u002Finstall\u002Fcopy semantics, package root, authored files | Package install\u002Fload |\n| Project integration | Hook file path, skill destination, MCP file, ownership\u002Frelocation, components-only projects, linked worktrees, root-checkout quirks | `sync`\u002F`verify` on a scratch project |\n| Foreign reads | Which other harnesses' project files does this harness load (hooks, skills, MCP, instructions), and in what form? Which other harnesses read this target's output? Is there an opt-out both sides honour? | A marker per location, both directions (ADR-0033) |\n| Packaging | Output paths, artifact modes, runtime bundle, shim execution, npm name\u002Fpublish, dependency install | Generated artifact + install |\n| Detection | Binary name, version command, PATH\u002FPATHEXT behavior, doctor output | `doctor` on each host |\n| Automation | Playback driver, pty driver, MCP driver, compaction\u002Fsubagent scenarios, drift driver, harness-watch lane | CI + local runs |\n| Documentation | Baseline, capture record, fixture provenance, generated support page, installation\u002Fproject\u002Ftesting\u002Fwatch docs | Review |\n\n### Budget the probe effort\n\nA one-session capture is not a harness integration. Plan for at least these\nindependent probes; combine only where one session genuinely decides several\nquestions:\n\n1. A passive lifecycle session with every event wired to a tee, both native\n   spellings where the harness accepts both.\n2. A tool-shape session per advertised tool family, including a failure path and\n   an MCP call.\n3. One effect-proof per write channel: block\u002Fdeny, input rewrite, output\n   rewrite, context injection, stop prevention, notification, session context.\n   Prove by a marker or a model-visible token, not by \"no error\".\n4. One shell-dialect probe per host configuration that could change the\n   interpreter, observed from the process that actually ran.\n5. A skill-policy probe with gated, ungated, explicit-true and explicit-false\n   controls; interactive slash invocation and headless prompt mode; project and\n   package routes; in-place and owned sources.\n6. MCP probes for every transport and both delivery routes, including\n   placeholders, environment\u002Fheader expansion, cwd and identity.\n7. One package install\u002Fload and one project `sync`\u002F`verify`, including a\n   components-only build.\n8. Negative controls: untrusted project, missing file, non-zero hook exit,\n   unsupported flag, unsupported transport.\n9. A foreign-read probe: put another harness's generated project output (for\n   example Claude's `.claude\u002Fsettings.json` hooks) in a scratch project and\n   observe whether this harness runs it, and the reverse.\n10. A version-boundary probe. Capture the exact build that supports a claim; do\n   not extend the claim to older builds without evidence.\n\nUse the cheapest transport that can decide the question. Loopback playback is\ndeterministic and free for tool calls; a pty is required when the harness has a\nseparate interactive path; a real model is justified only when model behavior\nis the question. A structured session log is stronger than a terminal screen:\na prompt can echo a token, and a model can read a file directly. Assert the\nstructured event that proves the effect.\n\n## Phase 1 — Capture reality\n\nFollow `harness-capture` for the mechanics and provenance. The integration-level\nrules:\n\n- Label provenance with the repository's ladder (`captured`, `live-probe`,\n  `router-log`, `schema-derived`, `type-derived`, `doc-derived`). Weaker evidence\n  is allowed only when it is named; never let a constructed or doc-derived shape\n  look captured.\n- Capture the harness, not the documentation. The existing adapters found\n  documented shapes that the binaries did not honor; docs are a hypothesis.\n- Keep raw captures under ignored `captured\u002F`; commit the drive scripts,\n  redacted fixtures, and a README that records question, method, observation,\n  consequence, and limits.\n- Capture both a positive and a negative control for every semantic claim.\n  \"Gated skill\" needs an ungated control; \"explicit allow\" needs an absent or\n  false control; \"block\" needs an allowed call.\n- Distinguish the model-facing tool from explicit user invocation. A `-p`\n  prompt naming `\u002Fskill` is not an interactive slash command; a `skill()` call\n  is not a user invocation. Use a pty for the interactive path and read the\n  session event log for `skill.invoked`\u002Ftool-call results.\n- Do not infer a shell dialect from the platform, a tool name, or a payload.\n  Observe the process that ran.\n- Do not infer MCP identity from a separator if either segment may contain that\n  separator, and do not split a lossy name. Use a stated identity or leave it\n  absent.\n- Capture declined surfaces too. An event with no portable meaning gets an\n  explicit declined input record and a sentence saying why; it is evidence of a\n  deliberate omission, not missing work.\n- Record a `validatedOn` record for every claim: exact version, date, method,\n  evidence path, and what the session established. A capture that skips this is\n  invisible to `inspect`, `doctor`, the generated support page, and the contract\n  suite.\n- No credential belongs in a capture. Use an isolated home and a disposable\n  scratch project; redact the capturing account path to `user` before committing\n  fixtures. Keep raw protocol traffic under ignored `captured\u002F`.\n- An event you cannot induce is **unmeasured**, not supported. Record the failed\n  procedure and do not advertise the event or add a capability cell until a real\n  payload exists. A decoder branch without a fixture is a latent path, not\n  evidence.\n- If a stop\u002Fprevent hook can loop without the documented guard, capture the\n  guard behavior and rate the cell `approximate` with the observed loop bound.\n  Do not infer a guard from the docs.\n\n## Phase 2 — Decide support honestly\n\n- Rate every capability and optional field. Absence means `unsupported`.\n- `exact` — the native mechanism matches the portable contract.\n- `emulated` — Hooknostic supplies the behavior the harness lacks, faithfully.\n- `approximate` — a useful native mechanism exists but its semantics differ.\n- `unsupported` — no honest translation; name the omission and the evidence.\n- Non-exact cells need a rationale. A partial mechanism is not silently\n  promoted; an unsupported cell with an honest reason is worth more than an\n  optimistic emulation.\n- Use a **deviation** when the harness departs from the specification for some\n  instances, a **degradation** when the projection cannot apply its own\n  translation to some items, and a **named omission** when the component cannot\n  reach the harness at all. Declare each with evidence from the profile.\n- Split profiles at the version where the evidence changes. Anchor the split to\n  the validation record, **not** `referenceVersion`: advancing the harness\n  metadata must not silently invalidate the build that was captured. A broad\n  target range resolves to the least-capable intersection; state that users may\n  need to narrow a range to keep a gate.\n- Every advertised event needs a fixture; every fixture needs an advertised\n  event. Every field a canonical fixture carries needs a rating, and every\n  rating needs a fixture. Every capability cell needs a scenario. Every shell\n  or file shape needs a fixture.\n- For skill invocation policy, decide all of: implicit selection, explicit\n  native invocation, interactive vs headless, package namespace, in-place\n  sources, and the native allow\u002Fdeny field. `false` must withhold implicit\n  selection **and** retain explicit user invocation; if a native flag does only\n  one half, approximate or unsupported is the honest rating. Name the acceptance\n  id for the shortfall. If a broad version range is not covered, keep the older\n  range unsupported rather than inheriting a newer claim.\n\n## Phase 3 — Build every surface\n\nWork through the adapter contract in `docs\u002Fadding-an-adapter.md`. The\nintegration-specific checklist:\n\n### Core adapter\n\n- `harness.ts`: display name, recommended range, fixture directory, reference\n  version.\n- `profile.ts`: fields, capability matrix, rationale, validation records,\n  version splits, deviations\u002Fdegradations\u002Fomissions.\n- `decode.ts`: tolerant reader, raw payload preserved, both native spellings,\n  no invented correlation ids, `DecodeError` for unmapped events.\n- `apply.ts`: strict writer; preserve native block distinctions and terminal\n  precedence; do not lower to a channel the harness ignores.\n- `toolmap.ts`: captured shell and file shapes only; a `ToolKind` entry for\n  every advertised tool; no guessed keys.\n- `shellCodec` \u002F `fileCodec` \u002F `classifyTool`: one table drives both directions;\n  pass the codecs into `dispatch()` so rewrites do not leave a stale view;\n  declare `shellDialect` and `mcpIdentity` with the evidence that supports them.\n- `generate.ts` \u002F `shim.ts`: command or module execution, argv\u002Fstdin\u002Fstdout,\n  timeout, fail-open behavior, `shimEntry`, `shimAliases`, `shimExecution`.\n  Shim value imports stay on SDK\u002Fruntime; never value-import core.\n- `detect()`: `detectCommandVersion` for PATH\u002FPATHEXT-safe version lookup.\n- `validateArtifacts()`: final-tree validation; throw means HN301, never a crash.\n- `hookRuntimePath()` and the projector's `packageRoot`: agree with the actual\n  output; the contract suite checks it.\n\n### Project delivery\n\n- `projectPaths`, `projectIntegration`, `projectComponents`,\n  `projectComponentProfiles`.\n- Hook file location, skill destination, MCP file, generated file ownership,\n  relocation\u002Frelinquish behavior, components-only projects, linked worktrees,\n  root-checkout paths.\n- `skillInvocationPolicy`: destination, alternate discovery directories, native\n  file projector, in-place detection, shared-directory composition.\n- Canonicalize symlink\u002Fjunction paths before deciding whether a skill is in\n  place. A source inside a native discovery directory is authored input: leave\n  it untouched and report `skill-policy-in-place`; never copy it onto itself or\n  claim its gate. A shared destination composes every reader's edits and\n  omissions, including readers whose `components.targets` excluded them.\n- Test `sync`, `verify`, components-only, authored files, and an existing copy.\n- Fence every foreign read (ADR-0033). A harness runs only its own target's\n  output: opt the reader out of another target's location with a mechanism\n  captured on both harnesses, or write somewhere it does not read, and pin it\n  with a cross-adapter test against the reader's captured rules. Never treat a\n  compatibility read as a delivery route. A shared standard location every\n  reader treats as its own (`.agents\u002Fskills`) is composed instead (ADR-0030).\n\n### Package \u002F Agent Plugin projection\n\n- `namespace`, `packageRoot`, profiles, `project`, `summary.components`,\n  `copiedPaths`, authored-hook merge, runtime package, artifact validation.\n- Carry every hook artifact through; a projection replaces the output.\n- Emit the harness's native format, not a portable document beside its\n  replacement.\n- Preserve authored hooks\u002Fcommands\u002Fagents\u002Frules\u002Fextensions; merge them when the\n  harness reads one document; error on a runtime-path collision.\n- Re-probe placeholders, package roots, data directories, install\u002Fcopy semantics,\n  and remote headers on the package route; a project fact is not package\n  evidence.\n- Apply skill policy after native overlays; preserve unrelated YAML, comments,\n  modes and body bytes; leave in-place sources untouched and report it.\n- `unsupported` means the component must not reach the harness, not merely that\n  it will be ignored. Drop it and record the omission.\n\n### Registration and automation\n\n- Register the adapter in `defaultAdapterRegistry()`.\n- Add the shim to `SHIMS` in `packages\u002Fcli\u002Fscripts\u002Fbundle.mjs`.\n- Add the adapter to `SUBJECTS` in `packages\u002Fcli\u002Fsrc\u002Fcoverage.test.ts`.\n- Add the harness lane to `scripts\u002Fharness-lanes.mjs`, the harness-watch workflow\n  matrix\u002Finstall\u002Fbootstrap cases, the playback driver, the drift driver, and\n  expected variants. Keep the default lane free and model-free; a paid\u002FLLM path\n  is opt-in and must say what it establishes.\n- Add `HOOKNOSTIC_PLAYBACK=\u003Cid>` scenarios for lifecycle, effects, MCP, pty,\n  compaction\u002Fsubagent where applicable, and the package\u002Fproject install. Use\n  `HOOKNOSTIC_PLAYBACK_VERSION` to verify a newer build without moving the\n  reference.\n- Add an opt-in `HOOKNOSTIC_SMOKE=\u003Cid>` test that proves one blocking and one\n  mutating effect with a real model; never run it in normal CI.\n- Update `docs\u002Fharness-watch.md` with the lane's cost and any install\u002Fbootstrap\n  quirk.\n- Automated watch records playback; it never moves `referenceVersion`,\n  `recommendedRange`, `fixtureDir`, or a profile range. Range extension is a\n  human decision backed by a real capture.\n\n## Phase 4 — Verify like the claim is false\n\n- Fixture decode\u002Fcanonical\u002Fapply\u002Fgolden round-trip.\n- Generation determinism, artifact validation, detection and package tests.\n- `describeAdapterContract`: events, fields, capabilities, shapes, scenarios,\n  profile metadata, provenance evidence.\n- Project `sync`\u002F`verify`, package projection, skill policy, MCP identity.\n- Real-harness playback with loopback: artifact discovery, block\u002Frewrite\n  markers, lifecycle, context, MCP, pty, package\u002Fproject. A skipped scenario is\n  inconclusive, not passing coverage.\n- Drift playback against committed fixture shapes.\n- Mutation-verify every new test: revert the fix or apply a mutant, watch the\n  test fail, restore. This repository has shipped tests that passed against\n  their own defect.\n- Full gates with explicit exit codes: `pnpm lint`, `pnpm format:check`,\n  `pnpm build`, `pnpm test`, `HOOKNOSTIC_PLAYBACK=\u003Cid> pnpm test`,\n  `node scripts\u002Fcheck-version-literals.mjs`.\n- Run `HOOKNOSTIC_SMOKE=\u003Cid>` only when the owner asks: it spends model credits\n  and can create harness trust entries. A green smoke run is evidence, not a\n  substitute for the free playback lane.\n- Rebuild committed example artifacts from the repository root if any source\n  affects them; otherwise state they are unaffected.\n- Regenerate `docs\u002Fharness-support.md` from the registry, never hand-edit it.\n\n## Phase 5 — Document and review\n\n- Baseline: the verified native surface and its sources.\n- Capture README: questions, method, observations, declined inputs, limits.\n- Fixture README: provenance per case and the redaction note.\n- Generated support page: regenerated.\n- Installation, project-integration, testing and harness-watch docs: update the\n  new harness's commands and limits.\n- ADRs: write or extend one only for a semantic-model, distribution or\n  normalization decision. Do not hide a policy change in code.\n- Review the diff independently. Fix concrete findings, re-run the gates, and\n  repeat until the remaining findings are explicitly accepted. A clean review\n  is not a substitute for a capture.\n\n## Definition of done\n\nDo not call the integration complete until all of these are true:\n\n- [ ] Every advertised event has a captured fixture; every fixture maps to an\n      advertised event; declined events are recorded.\n- [ ] Every field and capability is rated; non-exact cells have rationale and\n      evidence; every cell maps to a scenario.\n- [ ] Every shell\u002Ffile shape entry has a captured fixture; a tool without\n      evidence stays absent from the table. MCP identity is stated or absent.\n- [ ] Every advertised write\u002Fblock\u002Fcontext\u002Fstop\u002Fnotify channel is proven by\n      effect, with negative controls and terminal precedence.\n- [ ] Every foreign read is probed in both directions and fenced or reported\n      (ADR-0033); no target's output runs under another harness.\n- [ ] Both delivery routes, components-only, authored files, runtime paths and\n      package\u002Fproject install are tested, explicitly unsupported, or not\n      applicable with a reason.\n- [ ] Skill policy is decided for implicit, explicit, interactive, headless,\n      package, project and in-place paths; acceptance ids name every shortfall.\n- [ ] Version splits are anchored to validation records; a broad range's\n      least-capable behavior is documented; `validatedOn` records exist.\n- [ ] Playback, drift, harness-watch and smoke coverage are wired; a skipped\n      scenario is marked inconclusive, not counted as evidence.\n- [ ] Full gates pass with their own exit codes; generated docs and examples are\n      rebuilt; no generated file was hand-edited.\n- [ ] No downstream consumer is named; nothing is published; local distribution\n      remains `pnpm pack`.\n\n## Failure modes worth re-reading\n\n- **Docs are not the binary.** Every shipped adapter found a documented shape\n  the harness did not honor. Capture before implementing.\n- **`-p` is not interactive.** A prompt naming `\u002Fskill` is text; the model-facing\n  `skill()` tool is not user invocation. Use a pty and session logs for the\n  interactive path. The same split applies to permission prompts: a headless\n  session cannot ask a user.\n- **A native flag may enforce half a policy.** Capture explicit true\u002Ffalse and\n  implicit behavior separately; rate the half that is missing.\n- **The harness may report success for a skipped hook.** A missing error is not\n  evidence; observe the marker, rewrite, context injection, or model-visible\n  token.\n- **A prompt can echo the answer.** A terminal token check can pass without\n  loading the skill. Assert `skill.invoked`, a transformed skill-context\n  message, a tool-call result, or a marker file.\n- **A boundary tied to `referenceVersion` moves when metadata advances.**\n  Anchor a support split to the validation record that proved it.\n- **A known-broken older release is a range boundary.** Narrow the harness\n  range, capability profiles and projector profiles together; a package that\n  installs hooks which silently never run is worse than an honest unsupported\n  target.\n- **A package skill may be namespaced in the slash picker.** Capture the real\n  command, not the project skill name.\n- **A projection replaces the output.** Carry every hook artifact and merge\n  authored files; do not silently drop hooks or commands.\n- **A live plugin loader may copy nothing.** A package runtime dependency does not\n  appear by itself; bundle it or state the limitation.\n- **Harnesses read each other's files.** Copilot ran Claude's generated\n  `.claude\u002Fsettings.json` hook as a bare `node` and failed closed on every\n  shell call, in any project with Claude hooks. A target's tests pass alone;\n  probe the other harnesses in the same project.\n- **Placeholders differ by route.** A project MCP fact is not package evidence;\n  expansion in args, env, cwd and headers must be captured per route.\n- **Windows detection is PATHEXT, not `spawn`.** Use `detectCommandVersion`.\n- **Shims must not value-import core.** It drags esbuild into every artifact.\n- **A profile is a `Partial` record.** A typo in a capability id silently rates\n  nothing; the contract suite is the guard.\n- **Generated output is byte-gated.** Rebuild docs and examples from the repo\n  root; never hand-edit them.\n","# Harness integration\n\nSupport is a matrix of evidence-backed answers, not a compiling adapter. A\nharness is integrated when every surface it exposes is either implemented and\nproven, or explicitly declined with a reason and a provenance record. \"It\nbuilds\" and \"the docs say so\" are not completion criteria.\n\nThis skill is the orchestration and completeness layer. The implementation\nmechanics live in `docs\u002Fadding-an-adapter.md`; capture technique lives in\n`.agents\u002Fskills\u002Fharness-capture\u002FSKILL.md`; vocabulary changes live in\n`.agents\u002Fskills\u002Fextend-vocabulary\u002FSKILL.md`.\n\n## Required context\n\nRead before planning:\n\n- `AGENTS.md` and `CONTRIBUTING.md` — hard rules, scope, verification discipline.\n- `docs\u002Fadding-an-adapter.md` — the authoritative adapter workflow.\n- `docs\u002Fdesign.md` Appendix C — the definition of done.\n- `docs\u002Ftesting.md` — fixture replay, real-harness playback, pty and smoke layers.\n- `docs\u002Fharness-watch.md` — scheduled verification, drift, record, merge lanes.\n- Relevant ADRs in `docs\u002Fdecisions\u002F`: 0007 (tool shapes), 0008 (version facts),\n  0009 (rolling playback record), 0010 (scenario coverage), 0011 (Agent Plugin\n  projection), 0019 (deviations), 0020 (plugin root), 0021 (degradations),\n  0022 (shortfall policy), 0026 (file shapes), 0027 (optional fields),\n  0028 (skill directory), 0029 (shell dialect), 0030 (skill invocation policy),\n  0031 (named omissions), 0032 (MCP identity).\n\n## Phase 0 — Plan the integration\n\nBuild a surface matrix before writing code. The plan is a list of questions and\nthe evidence that will answer each one, not a list of files.\n\n### Choose the reference truth\n\n- Pick the harness id, binary or npm package, reference build, and fixture\n  directory name. Record them in `harness.ts`; every test and range derives from\n  that metadata.\n- If an older release is known broken, set the recommended range to the proven\n  boundary and propagate it to capability and projector profiles. Narrowing one\n  component while another still claims the broken range ships silent failures.\n- Record the platform matrix. A Windows-only capture does not establish Linux\n  shell behavior, and a headless capture does not establish an interactive\n  path. Plan a container or a second host where the harness ships one.\n- Decide the delivery modes: package, project, or both. A harness can have a\n  working package route and no project route, or the reverse. Do not assume one\n  from the other.\n- Decide whether the harness has a plugin\u002FAgent Plugins format, an in-process\n  extension model, or only command hooks. This determines the shim shape.\n\n### Surface inventory\n\nCopy this into the plan and fill every row with `capture`, `declined`, or `not\napplicable`. A blank row is an unowned risk.\n\n| Surface | Questions to answer | Evidence |\n| --- | --- | --- |\n| Process model | Command hook, in-process callback, imported module, long-lived server? What invokes it; what is stdin\u002Fstdout\u002Fargv; what is the timeout and failure mode? | Live harness session |\n| Hook envelope | Event names in every spelling; session\u002Fcwd\u002Ftimestamp\u002Fcorrelation fields; does the payload name its event? | Captured payload |\n| Lifecycle | Session start\u002Fend, resume, source, model, cwd, worktree, trust, folder-open behavior | Passive capture |\n| Prompt | User prompt before\u002Fafter transformation, output channels, block\u002Fcontext delivery | Passive + effect capture |\n| Tool before\u002Fafter | Native names, argument keys, result shapes, errors, parallel\u002Fbackground calls, MCP names | Tool-shape capture |\n| Effects | Block\u002Fdeny, ask\u002Fapproval, input rewrite, output rewrite, context injection, stop prevention, notification, session context | One effect-proof per channel |\n| Shell | Every shell tool advertised on each host; command\u002Fcwd keys; interpreter identity and dialect | Payload + process identity |\n| File tools | Read\u002Fwrite\u002Fedit\u002Fpatch\u002Fsearch shapes; path key or patch grammar; read-side limits | Captured tool calls |\n| Skills | Project and personal discovery; package discovery; body\u002Fpath expansion; frontmatter; explicit vs implicit invocation; native gate true\u002Ffalse; plugin namespace; in-place sources | Interactive TUI + `-p` controls |\n| MCP | stdio, Streamable HTTP, SSE; command\u002Fargs\u002Fenv\u002Fcwd; remote headers; placeholder expansion; startup; tool identity; runtime dependencies | Project and package probes |\n| Agent Plugins\u002Fplugin | Manifest, namespace, hooks, commands, agents, rules, LSP, client extensions, runtime package, load\u002Finstall\u002Fcopy semantics, package root, authored files | Package install\u002Fload |\n| Project integration | Hook file path, skill destination, MCP file, ownership\u002Frelocation, components-only projects, linked worktrees, root-checkout quirks | `sync`\u002F`verify` on a scratch project |\n| Foreign reads | Which other harnesses' project files does this harness load (hooks, skills, MCP, instructions), and in what form? Which other harnesses read this target's output? Is there an opt-out both sides honour? | A marker per location, both directions (ADR-0033) |\n| Packaging | Output paths, artifact modes, runtime bundle, shim execution, npm name\u002Fpublish, dependency install | Generated artifact + install |\n| Detection | Binary name, version command, PATH\u002FPATHEXT behavior, doctor output | `doctor` on each host |\n| Automation | Playback driver, pty driver, MCP driver, compaction\u002Fsubagent scenarios, drift driver, harness-watch lane | CI + local runs |\n| Documentation | Baseline, capture record, fixture provenance, generated support page, installation\u002Fproject\u002Ftesting\u002Fwatch docs | Review |\n\n### Budget the probe effort\n\nA one-session capture is not a harness integration. Plan for at least these\nindependent probes; combine only where one session genuinely decides several\nquestions:\n\n1. A passive lifecycle session with every event wired to a tee, both native\n   spellings where the harness accepts both.\n2. A tool-shape session per advertised tool family, including a failure path and\n   an MCP call.\n3. One effect-proof per write channel: block\u002Fdeny, input rewrite, output\n   rewrite, context injection, stop prevention, notification, session context.\n   Prove by a marker or a model-visible token, not by \"no error\".\n4. One shell-dialect probe per host configuration that could change the\n   interpreter, observed from the process that actually ran.\n5. A skill-policy probe with gated, ungated, explicit-true and explicit-false\n   controls; interactive slash invocation and headless prompt mode; project and\n   package routes; in-place and owned sources.\n6. MCP probes for every transport and both delivery routes, including\n   placeholders, environment\u002Fheader expansion, cwd and identity.\n7. One package install\u002Fload and one project `sync`\u002F`verify`, including a\n   components-only build.\n8. Negative controls: untrusted project, missing file, non-zero hook exit,\n   unsupported flag, unsupported transport.\n9. A foreign-read probe: put another harness's generated project output (for\n   example Claude's `.claude\u002Fsettings.json` hooks) in a scratch project and\n   observe whether this harness runs it, and the reverse.\n10. A version-boundary probe. Capture the exact build that supports a claim; do\n   not extend the claim to older builds without evidence.\n\nUse the cheapest transport that can decide the question. Loopback playback is\ndeterministic and free for tool calls; a pty is required when the harness has a\nseparate interactive path; a real model is justified only when model behavior\nis the question. A structured session log is stronger than a terminal screen:\na prompt can echo a token, and a model can read a file directly. Assert the\nstructured event that proves the effect.\n\n## Phase 1 — Capture reality\n\nFollow `harness-capture` for the mechanics and provenance. The integration-level\nrules:\n\n- Label provenance with the repository's ladder (`captured`, `live-probe`,\n  `router-log`, `schema-derived`, `type-derived`, `doc-derived`). Weaker evidence\n  is allowed only when it is named; never let a constructed or doc-derived shape\n  look captured.\n- Capture the harness, not the documentation. The existing adapters found\n  documented shapes that the binaries did not honor; docs are a hypothesis.\n- Keep raw captures under ignored `captured\u002F`; commit the drive scripts,\n  redacted fixtures, and a README that records question, method, observation,\n  consequence, and limits.\n- Capture both a positive and a negative control for every semantic claim.\n  \"Gated skill\" needs an ungated control; \"explicit allow\" needs an absent or\n  false control; \"block\" needs an allowed call.\n- Distinguish the model-facing tool from explicit user invocation. A `-p`\n  prompt naming `\u002Fskill` is not an interactive slash command; a `skill()` call\n  is not a user invocation. Use a pty for the interactive path and read the\n  session event log for `skill.invoked`\u002Ftool-call results.\n- Do not infer a shell dialect from the platform, a tool name, or a payload.\n  Observe the process that ran.\n- Do not infer MCP identity from a separator if either segment may contain that\n  separator, and do not split a lossy name. Use a stated identity or leave it\n  absent.\n- Capture declined surfaces too. An event with no portable meaning gets an\n  explicit declined input record and a sentence saying why; it is evidence of a\n  deliberate omission, not missing work.\n- Record a `validatedOn` record for every claim: exact version, date, method,\n  evidence path, and what the session established. A capture that skips this is\n  invisible to `inspect`, `doctor`, the generated support page, and the contract\n  suite.\n- No credential belongs in a capture. Use an isolated home and a disposable\n  scratch project; redact the capturing account path to `user` before committing\n  fixtures. Keep raw protocol traffic under ignored `captured\u002F`.\n- An event you cannot induce is **unmeasured**, not supported. Record the failed\n  procedure and do not advertise the event or add a capability cell until a real\n  payload exists. A decoder branch without a fixture is a latent path, not\n  evidence.\n- If a stop\u002Fprevent hook can loop without the documented guard, capture the\n  guard behavior and rate the cell `approximate` with the observed loop bound.\n  Do not infer a guard from the docs.\n\n## Phase 2 — Decide support honestly\n\n- Rate every capability and optional field. Absence means `unsupported`.\n- `exact` — the native mechanism matches the portable contract.\n- `emulated` — Hooknostic supplies the behavior the harness lacks, faithfully.\n- `approximate` — a useful native mechanism exists but its semantics differ.\n- `unsupported` — no honest translation; name the omission and the evidence.\n- Non-exact cells need a rationale. A partial mechanism is not silently\n  promoted; an unsupported cell with an honest reason is worth more than an\n  optimistic emulation.\n- Use a **deviation** when the harness departs from the specification for some\n  instances, a **degradation** when the projection cannot apply its own\n  translation to some items, and a **named omission** when the component cannot\n  reach the harness at all. Declare each with evidence from the profile.\n- Split profiles at the version where the evidence changes. Anchor the split to\n  the validation record, **not** `referenceVersion`: advancing the harness\n  metadata must not silently invalidate the build that was captured. A broad\n  target range resolves to the least-capable intersection; state that users may\n  need to narrow a range to keep a gate.\n- Every advertised event needs a fixture; every fixture needs an advertised\n  event. Every field a canonical fixture carries needs a rating, and every\n  rating needs a fixture. Every capability cell needs a scenario. Every shell\n  or file shape needs a fixture.\n- For skill invocation policy, decide all of: implicit selection, explicit\n  native invocation, interactive vs headless, package namespace, in-place\n  sources, and the native allow\u002Fdeny field. `false` must withhold implicit\n  selection **and** retain explicit user invocation; if a native flag does only\n  one half, approximate or unsupported is the honest rating. Name the acceptance\n  id for the shortfall. If a broad version range is not covered, keep the older\n  range unsupported rather than inheriting a newer claim.\n\n## Phase 3 — Build every surface\n\nWork through the adapter contract in `docs\u002Fadding-an-adapter.md`. The\nintegration-specific checklist:\n\n### Core adapter\n\n- `harness.ts`: display name, recommended range, fixture directory, reference\n  version.\n- `profile.ts`: fields, capability matrix, rationale, validation records,\n  version splits, deviations\u002Fdegradations\u002Fomissions.\n- `decode.ts`: tolerant reader, raw payload preserved, both native spellings,\n  no invented correlation ids, `DecodeError` for unmapped events.\n- `apply.ts`: strict writer; preserve native block distinctions and terminal\n  precedence; do not lower to a channel the harness ignores.\n- `toolmap.ts`: captured shell and file shapes only; a `ToolKind` entry for\n  every advertised tool; no guessed keys.\n- `shellCodec` \u002F `fileCodec` \u002F `classifyTool`: one table drives both directions;\n  pass the codecs into `dispatch()` so rewrites do not leave a stale view;\n  declare `shellDialect` and `mcpIdentity` with the evidence that supports them.\n- `generate.ts` \u002F `shim.ts`: command or module execution, argv\u002Fstdin\u002Fstdout,\n  timeout, fail-open behavior, `shimEntry`, `shimAliases`, `shimExecution`.\n  Shim value imports stay on SDK\u002Fruntime; never value-import core.\n- `detect()`: `detectCommandVersion` for PATH\u002FPATHEXT-safe version lookup.\n- `validateArtifacts()`: final-tree validation; throw means HN301, never a crash.\n- `hookRuntimePath()` and the projector's `packageRoot`: agree with the actual\n  output; the contract suite checks it.\n\n### Project delivery\n\n- `projectPaths`, `projectIntegration`, `projectComponents`,\n  `projectComponentProfiles`.\n- Hook file location, skill destination, MCP file, generated file ownership,\n  relocation\u002Frelinquish behavior, components-only projects, linked worktrees,\n  root-checkout paths.\n- `skillInvocationPolicy`: destination, alternate discovery directories, native\n  file projector, in-place detection, shared-directory composition.\n- Canonicalize symlink\u002Fjunction paths before deciding whether a skill is in\n  place. A source inside a native discovery directory is authored input: leave\n  it untouched and report `skill-policy-in-place`; never copy it onto itself or\n  claim its gate. A shared destination composes every reader's edits and\n  omissions, including readers whose `components.targets` excluded them.\n- Test `sync`, `verify`, components-only, authored files, and an existing copy.\n- Fence every foreign read (ADR-0033). A harness runs only its own target's\n  output: opt the reader out of another target's location with a mechanism\n  captured on both harnesses, or write somewhere it does not read, and pin it\n  with a cross-adapter test against the reader's captured rules. Never treat a\n  compatibility read as a delivery route. A shared standard location every\n  reader treats as its own (`.agents\u002Fskills`) is composed instead (ADR-0030).\n\n### Package \u002F Agent Plugin projection\n\n- `namespace`, `packageRoot`, profiles, `project`, `summary.components`,\n  `copiedPaths`, authored-hook merge, runtime package, artifact validation.\n- Carry every hook artifact through; a projection replaces the output.\n- Emit the harness's native format, not a portable document beside its\n  replacement.\n- Preserve authored hooks\u002Fcommands\u002Fagents\u002Frules\u002Fextensions; merge them when the\n  harness reads one document; error on a runtime-path collision.\n- Re-probe placeholders, package roots, data directories, install\u002Fcopy semantics,\n  and remote headers on the package route; a project fact is not package\n  evidence.\n- Apply skill policy after native overlays; preserve unrelated YAML, comments,\n  modes and body bytes; leave in-place sources untouched and report it.\n- `unsupported` means the component must not reach the harness, not merely that\n  it will be ignored. Drop it and record the omission.\n\n### Registration and automation\n\n- Register the adapter in `defaultAdapterRegistry()`.\n- Add the shim to `SHIMS` in `packages\u002Fcli\u002Fscripts\u002Fbundle.mjs`.\n- Add the adapter to `SUBJECTS` in `packages\u002Fcli\u002Fsrc\u002Fcoverage.test.ts`.\n- Add the harness lane to `scripts\u002Fharness-lanes.mjs`, the harness-watch workflow\n  matrix\u002Finstall\u002Fbootstrap cases, the playback driver, the drift driver, and\n  expected variants. Keep the default lane free and model-free; a paid\u002FLLM path\n  is opt-in and must say what it establishes.\n- Add `HOOKNOSTIC_PLAYBACK=\u003Cid>` scenarios for lifecycle, effects, MCP, pty,\n  compaction\u002Fsubagent where applicable, and the package\u002Fproject install. Use\n  `HOOKNOSTIC_PLAYBACK_VERSION` to verify a newer build without moving the\n  reference.\n- Add an opt-in `HOOKNOSTIC_SMOKE=\u003Cid>` test that proves one blocking and one\n  mutating effect with a real model; never run it in normal CI.\n- Update `docs\u002Fharness-watch.md` with the lane's cost and any install\u002Fbootstrap\n  quirk.\n- Automated watch records playback; it never moves `referenceVersion`,\n  `recommendedRange`, `fixtureDir`, or a profile range. Range extension is a\n  human decision backed by a real capture.\n\n## Phase 4 — Verify like the claim is false\n\n- Fixture decode\u002Fcanonical\u002Fapply\u002Fgolden round-trip.\n- Generation determinism, artifact validation, detection and package tests.\n- `describeAdapterContract`: events, fields, capabilities, shapes, scenarios,\n  profile metadata, provenance evidence.\n- Project `sync`\u002F`verify`, package projection, skill policy, MCP identity.\n- Real-harness playback with loopback: artifact discovery, block\u002Frewrite\n  markers, lifecycle, context, MCP, pty, package\u002Fproject. A skipped scenario is\n  inconclusive, not passing coverage.\n- Drift playback against committed fixture shapes.\n- Mutation-verify every new test: revert the fix or apply a mutant, watch the\n  test fail, restore. This repository has shipped tests that passed against\n  their own defect.\n- Full gates with explicit exit codes: `pnpm lint`, `pnpm format:check`,\n  `pnpm build`, `pnpm test`, `HOOKNOSTIC_PLAYBACK=\u003Cid> pnpm test`,\n  `node scripts\u002Fcheck-version-literals.mjs`.\n- Run `HOOKNOSTIC_SMOKE=\u003Cid>` only when the owner asks: it spends model credits\n  and can create harness trust entries. A green smoke run is evidence, not a\n  substitute for the free playback lane.\n- Rebuild committed example artifacts from the repository root if any source\n  affects them; otherwise state they are unaffected.\n- Regenerate `docs\u002Fharness-support.md` from the registry, never hand-edit it.\n\n## Phase 5 — Document and review\n\n- Baseline: the verified native surface and its sources.\n- Capture README: questions, method, observations, declined inputs, limits.\n- Fixture README: provenance per case and the redaction note.\n- Generated support page: regenerated.\n- Installation, project-integration, testing and harness-watch docs: update the\n  new harness's commands and limits.\n- ADRs: write or extend one only for a semantic-model, distribution or\n  normalization decision. Do not hide a policy change in code.\n- Review the diff independently. Fix concrete findings, re-run the gates, and\n  repeat until the remaining findings are explicitly accepted. A clean review\n  is not a substitute for a capture.\n\n## Definition of done\n\nDo not call the integration complete until all of these are true:\n\n- [ ] Every advertised event has a captured fixture; every fixture maps to an\n      advertised event; declined events are recorded.\n- [ ] Every field and capability is rated; non-exact cells have rationale and\n      evidence; every cell maps to a scenario.\n- [ ] Every shell\u002Ffile shape entry has a captured fixture; a tool without\n      evidence stays absent from the table. MCP identity is stated or absent.\n- [ ] Every advertised write\u002Fblock\u002Fcontext\u002Fstop\u002Fnotify channel is proven by\n      effect, with negative controls and terminal precedence.\n- [ ] Every foreign read is probed in both directions and fenced or reported\n      (ADR-0033); no target's output runs under another harness.\n- [ ] Both delivery routes, components-only, authored files, runtime paths and\n      package\u002Fproject install are tested, explicitly unsupported, or not\n      applicable with a reason.\n- [ ] Skill policy is decided for implicit, explicit, interactive, headless,\n      package, project and in-place paths; acceptance ids name every shortfall.\n- [ ] Version splits are anchored to validation records; a broad range's\n      least-capable behavior is documented; `validatedOn` records exist.\n- [ ] Playback, drift, harness-watch and smoke coverage are wired; a skipped\n      scenario is marked inconclusive, not counted as evidence.\n- [ ] Full gates pass with their own exit codes; generated docs and examples are\n      rebuilt; no generated file was hand-edited.\n- [ ] No downstream consumer is named; nothing is published; local distribution\n      remains `pnpm pack`.\n\n## Failure modes worth re-reading\n\n- **Docs are not the binary.** Every shipped adapter found a documented shape\n  the harness did not honor. Capture before implementing.\n- **`-p` is not interactive.** A prompt naming `\u002Fskill` is text; the model-facing\n  `skill()` tool is not user invocation. Use a pty and session logs for the\n  interactive path. The same split applies to permission prompts: a headless\n  session cannot ask a user.\n- **A native flag may enforce half a policy.** Capture explicit true\u002Ffalse and\n  implicit behavior separately; rate the half that is missing.\n- **The harness may report success for a skipped hook.** A missing error is not\n  evidence; observe the marker, rewrite, context injection, or model-visible\n  token.\n- **A prompt can echo the answer.** A terminal token check can pass without\n  loading the skill. Assert `skill.invoked`, a transformed skill-context\n  message, a tool-call result, or a marker file.\n- **A boundary tied to `referenceVersion` moves when metadata advances.**\n  Anchor a support split to the validation record that proved it.\n- **A known-broken older release is a range boundary.** Narrow the harness\n  range, capability profiles and projector profiles together; a package that\n  installs hooks which silently never run is worse than an honest unsupported\n  target.\n- **A package skill may be namespaced in the slash picker.** Capture the real\n  command, not the project skill name.\n- **A projection replaces the output.** Carry every hook artifact and merge\n  authored files; do not silently drop hooks or commands.\n- **A live plugin loader may copy nothing.** A package runtime dependency does not\n  appear by itself; bundle it or state the limitation.\n- **Harnesses read each other's files.** Copilot ran Claude's generated\n  `.claude\u002Fsettings.json` hook as a bare `node` and failed closed on every\n  shell call, in any project with Claude hooks. A target's tests pass alone;\n  probe the other harnesses in the same project.\n- **Placeholders differ by route.** A project MCP fact is not package evidence;\n  expansion in args, env, cwd and headers must be captured per route.\n- **Windows detection is PATHEXT, not `spawn`.** Use `detectCommandVersion`.\n- **Shims must not value-import core.** It drags esbuild into every artifact.\n- **A profile is a `Partial` record.** A typo in a capability id silently rates\n  nothing; the contract suite is the guard.\n- **Generated output is byte-gated.** Rebuild docs and examples from the repo\n  root; never hand-edit them.\n",{},"master","b8669236433862d74703a9a146bf1373f9b0b209",".agents\u002Fskills\u002Fharness-integration",0,"https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic","https:\u002F\u002Fgithub.com\u002Froryeckel","https:\u002F\u002F2labz.com","I write a lot of software and delegate the rest to LLMs","2026-11-15 04:44:54",{"source":7,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add roryeckel\u002Fhooknostic --skill harness-integration",{"tool":38,"command":39},"bunx","bunx skills add roryeckel\u002Fhooknostic --skill harness-integration",{"tool":41,"command":42},"pnpm","pnpm dlx skills add roryeckel\u002Fhooknostic --skill harness-integration","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fharness-integration\u002FSKILL.md",true,[],[47,54,59,64,69,76],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":18},"01M7AXANW27HCE8N4ZD2H8GZG5","sample","Explain the synthetic sample project.","https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic\u002Fblob\u002Fmaster\u002Fexamples\u002Flocal-project\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":55,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":56,"categories":57,"keywords":58,"publishedAt":18},"01M7AXAKKRACEVMN7CZ44ZC6A1","https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic\u002Fblob\u002Fmaster\u002Fexamples\u002Flocal-project\u002F.pi\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":60,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":61,"categories":62,"keywords":63,"publishedAt":18},"01M7AXAJEYX4DKRCRBSNPM7J09","https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic\u002Fblob\u002Fmaster\u002Fexamples\u002Flocal-project\u002F.claude\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":65,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":18},"01M7AXAFPMXR8EJM4D2Z1WKBH2","https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic\u002Fblob\u002Fmaster\u002Fexamples\u002Flocal-project\u002F.agents\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":18},"01M7AXADDP9ASKQBVG8T8KE3XV","greet","Greet the user by name and summarize the repo state.","https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic\u002Fblob\u002Fmaster\u002Fexamples\u002Fagent-plugin\u002Fskills\u002Fgreet\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":18},"01M7AXAC2V6M2XGEEBAPARQA4R","where","Reports where this skill lives. Use when asked where the probe skill is.","https:\u002F\u002Fgithub.com\u002Froryeckel\u002Fhooknostic\u002Fblob\u002Fmaster\u002F.capture\u002Fskill-directory\u002FSKILL.md",[],[]]