[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77KSNBGS4N933X65DFGSBX0":3,"skill-same:01M77KSNBGS4N933X65DFGSBX0":144},{"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":30,"commitSha":31,"skillDirectory":32,"forksCount":33,"repositoryHtmlUrl":34,"repositoryHomepageUrl":6,"ownerHtmlUrl":35,"ownerBlogUrl":36,"ownerDescription":37,"updatedAt":38,"install":39,"resources":51},"01M77KSNBGS4N933X65DFGSBX0","audit-pass",null,"Run one coordinated, resumable audit pass over a repo: a three-scope inventory, then each owning plugin's checks lane by lane, presence-gated, with findings persisted per lane and one human gate for the whole pass. Read-only unless --fix. Use when: 'audit pass', 'run one pass over this repo', 'coordinate the audit skills', 'audit this repo end to end', 'resume the audit', 'one reconciled findings report', 'sweep all three scopes', or a release needs one diffable findings artifact.","melodic-software\u002Fclaude-code-plugins","melodic-software","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F58273638?v=4","01M77JVQTXNGKCE4MDFY2BC936","Melodic Software","01M77JVQVN9QDWKAH6W03ANP76",30,"MIT","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Run one coordinated, resumable audit pass over a repo: a three-scope inventory, then each owning plugin's checks lane by lane, presence-gated, with findings persisted per lane and one human gate for the whole pass. Read-only unless --fix. Use when: 'audit pass', 'run one pass over this repo', 'coordinate the audit skills', 'audit this repo end to end', 'resume the audit', 'one reconciled findings report', 'sweep all three scopes', or a release needs one diffable findings artifact.\"\nargument-hint: \"[target] [--fix] [--opinion] [--postures] [--lanes \u003Clist>] [--resume] [--report-to \u003Cdir>]\"\nuser-invocable: true\ndisable-model-invocation: false\ndisallowed-tools: Edit, NotebookEdit\nmetadata:\n  workflow-stage: anytime\n  summary: Run one coordinated, resumable audit pass over a repo with a single human gate\n---\n\n## Purpose\n\nOne bounded, ordered pass over a target repository's Claude Code **instruction** surface, coordinated\nacross three scopes and resumable mid-run. It **adds no criteria of its own**: every check is\ndelegated to the plugin that owns it. What it contributes is what invoking those skills by hand\nyields none of: a three-scope inventory before any check runs, a run-time-derived exclusion set,\nstable finding identity, suppression memory, incremental persistence and resume, and one human gate\nfor the pass. All of it is specified in [reference\u002Frun-contract.md](reference\u002Frun-contract.md).\n\n## Reference index. Load on demand\n\n| File | Load when |\n|---|---|\n| [reference\u002Fterms.md](reference\u002Fterms.md) | Before any other contract file. Every one of them uses run, target, lane, scan set, live surface and live surface set without redefining them. |\n| [reference\u002Ffinding-identity.md](reference\u002Ffinding-identity.md) | Emitting, comparing, or suppressing a finding: the `(check, claim, sites)` tuple, `surface`, `anchor`, and the derived `finding_id`. |\n| [reference\u002Frun-contract.md](reference\u002Frun-contract.md) | Deciding which contract file owns a rule, or resolving a `§N` cross-reference from inside one. |\n| [reference\u002Farguments.md](reference\u002Farguments.md) | A run is invoked with an argument whose precedence, default, or refusal is not settled by the `Arguments` table. |\n| [reference\u002Fexclusion-set.md](reference\u002Fexclusion-set.md) | Deriving the scan set, or justifying why a path was skipped. |\n| [reference\u002Frun-state-and-resumability.md](reference\u002Frun-state-and-resumability.md) | Starting, leasing, or resuming a run, and diagnosing a lock that outlived its holder. |\n| [reference\u002Freport-location-and-schema.md](reference\u002Freport-location-and-schema.md) | Writing the report, or judging what `--report-to` may target. |\n| [reference\u002Fdeterminism-tiers.md](reference\u002Fdeterminism-tiers.md) | Assigning a finding's tier, or running the self-check's comparison. |\n| [reference\u002Fsuppression.md](reference\u002Fsuppression.md) | Reading or writing the target's suppression record (`harness-config.yaml`). |\n| [reference\u002Fdoctor-handoff.md](reference\u002Fdoctor-handoff.md) | Reaching the `\u002Fdoctor` handoff, whether it is present or absent. |\n| [reference\u002Fliveness-sources.md](reference\u002Fliveness-sources.md) | Inventorying Phase 1 liveness: which sources this run can reach, what each one establishes, and the three losses it names in `skipped`. |\n| [reference\u002Flane-catalog.md](reference\u002Flane-catalog.md) | Dispatching Phase 3, or resolving what a `--lanes` id covers: every lane, what each delegated skill owns, what is not dispatched, and the substrate probe. |\n| [reference\u002Fretired-conventions-sweep.md](reference\u002Fretired-conventions-sweep.md) | Running the one script lane: how installed manifests are discovered, what an undiscoverable manifest set reports, and how a helper exit maps to a finding. |\n\n## Read-only contract, and where mutation can reach\n\nBare invocation reads and reports. `--fix` is the only mutation path, and it is bounded by scope:\n\n| Scope | Posture under `--fix` |\n|---|---|\n| Managed policy | **Never remediated.** Read-only in every mode. A finding here reports \"conflicts with org policy at `\u003Cpath>`\" and proposes no edit to either side; seeking a policy exception is an organizational decision, not a linting one. |\n| User | **Routed as a recommendation, never edited in place.** A user-scope tree is commonly managed by a dotfiles manager, so an in-place edit is drift the operator's own sync path will fight. |\n| Project | The only editable scope, per-finding confirmed. |\n\nThe frontmatter narrows the surface: `disallowed-tools: Edit, NotebookEdit` takes both editing\ntools away for a span the skills page bounds, so the read-and-report contract is instruction-held\nwith a narrowed accident surface, never an enforced guarantee. `Write` is kept, since run state and\nthe report persist under `${CLAUDE_PLUGIN_DATA}`, and it can mutate a file, so never describe the\nrestriction to an operator as mechanical. `audit-instructions` states the same posture.\n\n- **Pointer**: when deciding how long the restriction lasts, fetch [skills: frontmatter reference](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fskills#frontmatter-reference) live.\n\n**`scripts\u002Frun-state.sh` writes under that same plugin data directory and nowhere else**, never\ninside a target repository. It takes the data directory as an argument rather than discovering one,\nand validates both path segments it contributes: `lib\u002Fstate-key.sh` refuses a remote URL that would\nbecome traversing directory components, and a `--run-id` outside `[A-Za-z0-9][A-Za-z0-9_.-]*` is\nrefused here.\n\n**The run mechanics are scripts, not prose to re-derive per run.** Four of them, each with tests\nbeside it, and each is the *only* correct source for what it computes: a mechanic described in prose\nand performed by the model is a different mechanic per run, which is precisely what makes a\ndeterminism property unfalsifiable.\n\n| Script | Owns |\n|---|---|\n| `scripts\u002Frun-state.sh` | The run directory, the lease and its verdict, and the append-only partial. |\n| `scripts\u002Fstate-digest.sh` | `state-digest\u002Fv1` and `input-digest\u002Fv1`: the comparability input and every lane's resume digest. |\n| `scripts\u002Ffinding-identity.sh` | `anchor\u002Fv1` normalization, `finding_id\u002Fv1`, `group\u002Fv1`, and `guarded-append`: the guard, on the append path rather than beside it, so no record reaches the partial without passing it. |\n| `scripts\u002Fassemble.sh` | Partial-to-report assembly, and the `report.md` rendering of the assembled document. |\n\nThey read and write nothing inside the target: the digest scripts hash files the run names and print\na hex string, and the rest write only under the run directory.\n\n## Scope boundary (route out)\n\n- **One instruction surface against the model-capability catalog** → `\u002Fharness-config:audit-instructions`\n  directly. This pass dispatches that skill by invoking it via the Skill tool; it does not\n  re-answer it.\n- **Config-file correctness** is a delegated lane, not a route-out: this pass invokes\n  `\u002Fharness-config:audit` and appends the findings rows it persists (Phase 3).\n- **Grant portability** is a delegated lane at its `frontmatter` scope only (Phase 3). Its\n  settings-file scope routes out to `\u002Fharness-config:audit-permission-grants` directly, because the\n  permission plane as it is in effect is `audit-permission-state`'s lane here.\n- **Automation landscape** → `\u002Fharness-config:audit-automation-gaps`. Not in this pass's surface set:\n  it proposes automation that does not exist, so it reports over no inventoried surface.\n\n## Arguments\n\nParse `$ARGUMENTS`. The names and their one-line meanings:\n\n| Argument | Meaning |\n|---|---|\n| `target` | The git repository to audit. Defaults to the active project root; a `target` resolving anywhere else is refused non-zero. |\n| `--fix` | The explicit mutation override. Absent, the pass writes nothing into the target. |\n| `--opinion` | Run the `OPINION`-tier checks the delegated catalogs declare default-off. |\n| `--postures` | Dispatch the opt-in posture lane. Off by default: it is the additive catalog and the largest fan-out in the pass. |\n| `--lanes \u003Clist>` | Run only the named lanes, comma-separated. Narrows dispatch, never the inventory; an unknown lane id is refused non-zero. |\n| `--resume` | Resume the most recent incomplete run for this target's state key. |\n| `--report-to \u003Cdir>` | Redirect **both** report artifacts into that directory, subject to the destination gate. A directory, never a file path. |\n\nFlag semantics, precedence between them, and what `--fix` and `--resume` change about the phases\nbelow are in [reference\u002Farguments.md](reference\u002Farguments.md). Read it when a run is invoked with\nmore than one flag, or with any flag whose effect on a later phase you are about to assume.\n\n## Phase 0: Resolve, key, lock\n\nResolve the target root, compute the state key, and take the lock posture for the mode. Read-only\nruns take no lock and run concurrently; an applying run takes an exclusive advisory lock and refuses\nrather than queues. All specified in\n[reference\u002Frun-state-and-resumability.md](reference\u002Frun-state-and-resumability.md). With `--resume`,\nread the lease, then read the partial's lane records and carry forward every lane whose input digest\nis unchanged.\n\n**Do not derive the run directory or hand-write the lease.** `scripts\u002Frun-state.sh` does both, which\nis what makes this phase a mechanism rather than a description of one:\n\n```bash\nS=\"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Faudit-pass\u002Fscripts\u002Frun-state.sh\"\nD=\"${CLAUDE_PLUGIN_DATA}\"\nbash \"$S\" paths --plugin-data \"$D\" --run-id \"\u003Crun-id>\"\nbash \"$S\" lease acquire --run-dir \"\u003Crun-dir>\" --run-id \"\u003Crun-id>\" --plugin-data \"$D\"\n```\n\n`paths` derives `\u003Cplugin-data>\u002Fruns\u002F\u003Cstate-key>\u002F\u003Crun-id>` through the plugin's own `lib\u002Fstate-key.sh`,\nthe library whose header records the keying scheme. Pass `--plugin-data` explicitly with the `${CLAUDE_PLUGIN_DATA}` value from this text; never rely on the shell to expand it (Pointer: fetch [manifest reference: where each variable resolves](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fplugins\u002Fmanifest-reference#where-each-variable-resolves) live).\n`acquire` takes it too, and refuses a `--run-dir` that is not under\n`\u003Cplugin-data>\u002Fruns\u002F`. It is the only command that *creates* a directory, so it is where the write\ntree is pinned; a wrong or invented run dir would otherwise be created and written into.\n\n**`--resume` never attaches to a run that is still going.** Concurrent read-only runs are safe\nbecause each owns its own partial artifact; resume is the one operation that reaches into *another*\nrun's artifact, so the no-lock policy that makes concurrency safe is exactly what leaves resume\nunable to tell a live run from an interrupted one. Both would then append lane attempts and\nterminating records to one file, and highest-terminated-attempt assembly becomes race-dependent,\nthe interruption-tolerance mechanism producing a report neither run performed.\n\nSo every active run, read-only included, maintains a **lease**, and `--resume` reads it before it\nreads the partial. `run-state.sh lease classify --run-dir \u003Crun-dir>` prints the verdict:\na **live** lease means the run is still going, and resume exits non-zero naming the run id rather\nthan attaching; a **stale** lease means the run was interrupted and its artifact is resumable; a\n`released` tombstone is resumable immediately; `missing` means there is nothing to attach to. The\nlease is not a lock: it excludes nothing, blocks no concurrent read-only run, and grants no\nexclusivity; it answers the one question resume has to ask.\n\nRefresh it at every lane's persistence point (`lease heartbeat`) and write the tombstone on a clean\nexit (`lease release`). The full specification, covering path, contents, the two-sided liveness window, and an\nexplicit statement of **which clauses the script enforces and which remain the run's own discipline**,\nis in [reference\u002Frun-state-and-resumability.md](reference\u002Frun-state-and-resumability.md) §3. Read\nthat split before relying on any of it: a clause the script does not enforce is the run's own\ndiscipline, not a guarantee.\n\n**The scan baseline is captured after the inventory is frozen and before any lane reads.** The\ndigest spans every inventoried scope, so it cannot be computed before Phase 1 has produced that\ninventory. Taking it at the top of Phase 0 would either omit the user and managed surfaces, which\nare exactly the mid-run external edits the gate exists to detect, or force an unspecified second\ninventory. So Phase 0 resolves, keys, and locks; Phase 1 freezes the inventory; the **scan baseline**,\nthe target's HEAD commit and the run's state digest, is taken at that boundary, and the matching\n**audit endpoint** capture is taken when the last lane completes, before any Phase 5 mutation.\n\nBaseline to endpoint is therefore exactly the window in which lanes read, which is what the\ndeterminism gate is a claim about. A run that never measures it cannot claim it held. The digest\npairs each path with a hash of its current content, because a *count* holds still while a dirty\nfile's contents change underneath the run.\n\n## Phase 1: Three-scope inventory, before any check\n\n**Nothing is checked until all three scopes are inventoried.** A project-only inventory cannot see a\nproject-versus-user conflict, so a fix from one would be applied against half the picture.\nNative-first: the filesystem walk produces a **candidate** set, never the answer.\n\n**Liveness sources are the ones this run can actually reach, and a source it cannot reach is named\nrather than required.** Which sources this run can hold, which are out of reach and why, what each\none establishes, and the three coverage losses named in `skipped` rather than absorbed, are in\n[reference\u002Fliveness-sources.md](reference\u002Fliveness-sources.md). Read it before recording any\nliveness basis.\n\n**A liveness claim carries its basis, and the report marks a single-sourced one.** A single-sourced\ninventory is usable; silently presenting it as a multi-source result is the under-coverage-reads-as-\nclean failure the rule exists to prevent. The liveness basis records which sources were live, so a\nrun with an `InstructionsLoaded` producer and a run without are **not comparable** and cannot fail\nP1 against each other.\n\nRecord, per scope, every surface found **and every surface skipped with its reason**. The inventory\nis a reported derived-tier artifact, so a surface that silently drops out of scope between two runs\nfails the determinism property rather than looking like an improvement.\n\n**Output styles are the case a walk alone cannot get right.** A style changes the instructions\nClaude receives, and the live style can differ from the one the user's settings name, so a walk\nplus settings can report a selection as live when it is not. Report the resolved live style and\nwhat resolved it.\n\n- **Pointer**: when resolving the live style and what it changes, fetch [output styles: how output styles work](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Foutput-styles#how-output-styles-work) and [create a custom output style](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Foutput-styles#create-a-custom-output-style) live.\n\n**Shadowed definitions fall out of this inventory, not out of a catalog.** Where skills, subagents\nor MCP servers share a name across scopes, the pass compares names across the precedence order\nread live, deterministic, model-free and derived tier, and reports at `info` in its own section\nnaming the live definition and the shadowed one, never as a conflict finding or merged into one.\nWhen deciding which same-named definition is live, fetch [skills: resolve skills that share a name](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fskills#resolve-skills-that-share-a-name), [subagents: choose the subagent scope](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#choose-the-subagent-scope) and [MCP: scope hierarchy and precedence](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmcp#scope-hierarchy-and-precedence) live.\n\n## Phase 2: Derive the exclusion set\n\n**Derived at run time from the target's own state. Never transcribed, and no count of any class is\never carried in this skill.** A count written down is wrong on the next commit and wrong in every\nother repository. Four classes: registered byte-identical cluster copies, vendored upstream\nmaterializations, worktrees, and the pass's own artifacts. Each class's derivation, its fallback on a\nfailed read, and the hard error when a suppression targets an excluded path are in\n[reference\u002Fexclusion-set.md](reference\u002Fexclusion-set.md).\n\n## Phase 3: Lanes and dispatch\n\n**A lane is one delegated invocation at the finest filter that skill's own interface accepts**, the\ngranularity the pass can actually dispatch, since it invokes skills and never reaches inside one. A\nfiner lane is unbuildable, not merely inconvenient: per-lane persistence, input digests, and\nselective resume all key on something the pass can re-invoke on its own.\n\nSo the split falls out of the delegated interfaces rather than being asserted over them: a skill\ntaking a **surface-class filter** yields one lane per filter value it is given, and a skill taking\nonly an action verb yields **one lane** covering everything it audits. Where a skill runs its whole\ncatalog per invocation, the lane carries that whole catalog; the pass never splits a catalog it\ncannot address. Extending a delegated interface to accept a finer filter is a change to that skill,\nand until it lands the lane stays at the coarser grain.\n\n**What earns a lane, stated as a criterion rather than a list.** A skill is dispatched when it\nproduces findings **over surfaces the Phase 1 inventory already enumerates**, keyed to a site this\npass can anchor. That is what makes its rows joinable with the inventory, suppressible under this\npass's identity, and comparable across runs. `harness-config:audit-automation-gaps` fails the\ncriterion in both halves and is routed out: it is forward-looking, proposing automation that does not\nexist, so it has no inventoried surface to report over and no site to anchor a finding at.\n\n**Lane ids are what `--lanes` names**, one per lane in\n[reference\u002Flane-catalog.md](reference\u002Flane-catalog.md), plus one per surface-class value for the\ninstruction catalog: `instructions:\u003Cclass>`, `conflicts`, `permission-state`, `permission-grants`,\n`config`, `memory`, `retirements`, and `postures`. A `--lanes` value naming an id outside that set is\nrefused non-zero.\n\nDispatch, in inventory order, with every skill invoked via the Skill tool, each invocation\npresence-gated with its fallback stated. Every lane, what each delegated skill owns, why each is\nexactly one lane (or one per surface-class value), what is deliberately left undispatched, and the\nsubstrate probe that decides whether a lane can spawn its own verifier, are in\n[reference\u002Flane-catalog.md](reference\u002Flane-catalog.md). Load it before dispatching a lane, before\nanswering what a `--lanes` id covers, and before assuming a lane can verify itself.\n\nPersist each lane's findings to the partial artifact **as that lane completes**, never buffered to\nthe end. A lane is complete when its terminating record is in the partial, and every record carries\nits attempt id so an abandoned re-attempt is discardable rather than merely older. The write is one\ncall per record, and it goes through the guard rather than past it:\n\n```bash\nFI=\"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Faudit-pass\u002Fscripts\u002Ffinding-identity.sh\"\nbash \"$FI\" guarded-append --run-dir \"\u003Crun-dir>\" --record '\u003Cjson-line>' --epoch \"\u003Cheld>\"\n```\n\n`guarded-append` runs `validate-record` and appends only if it passes, so **no record reaches the\npartial without clearing §1**; `run-state.sh partial append` is what it calls and is never the call a\nlane makes itself. **Exit 4 is the guard's refusal and nothing was written**: a null identity, three\nor more sites, two sites with no pairwise declaration, an anchor outside the grammar, or a\n`finding_id\u002Fv1` disagreeing with its constituents. Fix the lane, never the record. The lease is\nrefreshed at the same boundary; a lease must exist, so a record resume could not attribute to a\nlive-or-abandoned run is never written. **Pass the epoch you hold**: the filename is the writer's\nepoch, not whatever the lease now carries, which keeps a fenced writer's rows out of its adopter's\nfile. The record is also validated as well-formed single-line JSON rather than sniffed by its first\ncharacter, because a malformed row here is permanent and resume is its only reader.\n\n**Exit 3 means FENCED: stop this run.** The record was written safely to your own epoch file, but the\nlease has moved on and another run has adopted the artifact, so continuing dispatches lanes whose\noutput nothing will assemble. Stop and report the run as superseded. Never retry the append, and never\nread that exit as transient.\n\n**The lane count is bounded by the delegated interfaces, not chosen here**, at one per scope value the\ninstruction catalog accepts plus one each for the permission plane, grants, config, the memory layer,\nthe retirements sweep, and the opt-in posture lane, so it is a handful, and a per-run\ndispatch ceiling would never bind. What is *not* bounded here is the fan-out inside a lane: the\ndelegated catalogs spawn their own subagents, and this pass cannot reach inside one to cap it. So cap\nconcurrency at 3–5 lanes and let incremental persistence carry the rest. It is what degrades a blown\nsession ceiling into a resumed run.\n\nThe guarded-append call above bounds nothing inside a lane; what it buys is that an intra-lane\noverrun costs the lanes still running rather than the whole pass.\n\n## Phase 4: The `\u002Fdoctor` handoff\n\nThis pass builds no replacement for `\u002Fdoctor` and hands the instruction-file audit to `\u002Fdoctor\nprompt-audit` rather than running it; for what each owns, fetch [memory: audit your instruction files](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmemory#audit-your-instruction-files)\nlive. **Both are\noperator-run, so neither is dispatched.** Their records (version floors, write posture),\nwhat the presence check verifies versus probes, and the absence classification are in\n[reference\u002Fdoctor-handoff.md](reference\u002Fdoctor-handoff.md).\nWhen `\u002Fdoctor` is absent, name it as the missing capability and state what goes unchecked.\n\n**Phase 4 records the handoff; it does not stop the pass.** Halting here would mean a `--fix` run\nnever reaches Phase 5 and no run reaches the Phase 6 report, so an optional collaborator would\ncancel the coordinated pass that is this skill's purpose. `reference\u002Fdoctor-handoff.md` says to\nfinish the pass's own phases first.\n\nSo Phase 4 opens the `delegated` lane, records the instruction in the report, and continues. Phases 5\nand 6 run normally, and the assembled report carries the handoff as an outstanding item with its lane\nmarked `open`, routed and not yet returned. The operator runs `\u002Fdoctor` when they choose; a later\n`--resume` closes that lane by re-prompt rather than re-scan, because nothing about it needs the\nsweep to run again.\n\n**`open` is an assembly terminator, not a completion.** The two are distinct and conflating them\nwould have made the promised resume impossible: §7 needs a terminating record to assemble a report at\nall, while §5 skips any lane whose state is complete and whose digest is unchanged, so a lane that\nwas both terminated *and* complete would be carried forward untouched on every resume, and the\noutstanding handoff would never close. So the record terminates the attempt for assembly and marks\nthe lane's state **incomplete**. `--resume` therefore re-runs it, which for a delegated lane means\nre-prompting rather than re-scanning. `handed-back` and `declined` are completions; only `open` is\nnot.\n\n**That instruction to the operator is only true if the terminating record is actually written.**\n`--resume` reads the partial, not the report, so a report telling the operator to come back with\n`--resume` against a partial nothing wrote is a false instruction in the one artifact they act on. So\nthe `open` terminator is written at the moment Phase 4 records the handoff, never deferred to Phase 6\nassembly, which is exactly where a run that does not reach Phase 6 loses it. It goes through\n`guarded-append` like every other record: the guard short-circuits on anything that is not a\n`finding`, so guarding a terminator costs nothing, and **one write command for the whole skill** is\nwhat keeps Phase 3's rule from reading as advice a coordinator may opt out of.\n\n## Phase 5: Apply, only under `--fix`\n\nPer-finding confirmation, project scope only, bounded by the scope table under \"Read-only contract, and where mutation can reach\". Refuse and name the\ncanonical source if a fix or a suppression would write into a derived-exclusion path.\n\nThe apply-verify step judges work this same run produced, so it is delegated: hand the applied diff\nand the finding it claims to resolve, the artifact rather than this run's reasoning, to a cross-vendor\nadvisor when one is installed and set up (the OpenAI Codex plugin, say, invoked per its own docs),\nwith a **fresh-context (non-fork) subagent** as the stated fallback. Resolve the `verification` level once per run with `bash \"${CLAUDE_PLUGIN_ROOT}\u002Flib\u002Fread-verification.sh\" --plugin harness-config --skill audit-pass` and copy its line into the report header ([levels](https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fdocs\u002Fcontracts\u002Fverification-level\u002FREADME.md)). Under `light`, and `off`, which acts as `light` here, neither this apply-verify dispatch nor Phase 6's main-session verification runs and nothing is verified inline in their place: record `skipped` for the reason `verification: skipped (\u003Clevel>)` and mark findings and fixes `(unverified)`, as \"When dispatch is unavailable\" below says. A delegated lane applies its own skill's level.\n\n### When dispatch is unavailable\n\nThe apply-verify step and delegated lanes that mandate subagent dispatch **require** that dispatch.\nWhen the Agent tool is blocked, unavailable, or the probed depth budget does not permit the spawn a\nstep needs:\n\n1. **Record per-lane verification mode** in the lane's terminating record and the assembled report (`verified` |\n   `inline` | `skipped`) for every lane that mandates independent verification.\n2. **Mark unverified findings.** Proposals or applied fixes that did not receive an independent\n   verifier carry an `(unverified)` marker and are never presented as resolved.\n3. **Do not silently complete.** The `skipped` section and report header name dispatch\n   unavailability, and the probed depth budget, when either prevented a mandated verification phase.\n4. **The main-session pass is the one that survives a lowered depth budget**, because it spawns from\n   the top of the session rather than from inside a lane. A run that could verify no lane inline can\n   usually still verify the assembled set, so `inline` per lane and a completed Phase 6 verification\n   is the ordinary shape on a depth-1 host, not a degraded one.\n\n## Phase 6: Assemble, verify the assembled set, report\n\n**Assemble first.** An append-only `findings.partial.\u003Cowner_epoch>.jsonl` during the run becomes\n`findings.json` here, epoch-scoped so a fenced writer cannot interleave into its adopter's file, with\nassembly reading only the highest epoch present and, per lane, the highest-ordinal attempt carrying a\nterminating record. `scripts\u002Fassemble.sh` performs it. Both schemas, the identity-versus-presentation\nfield split, and the sections are in\n[reference\u002Freport-location-and-schema.md](reference\u002Freport-location-and-schema.md); the tier\nmemberships are in [reference\u002Fdeterminism-tiers.md](reference\u002Fdeterminism-tiers.md). Tiers stay in\nseparate sections because their guarantees differ.\n\n### Main-session verification, after assembly\n\n**A verification pass runs here, in the main session, on the assembled set, and its justification is\nscope rather than inability.** A lane can only ever verify its own findings; several relations exist\nonly once every lane's output is in one document, and nothing inside a lane can see them:\n\n- a finding two catalogs both raised at one site, which is the overlap note's whole subject;\n- a finding contradicting a suppression the record carries for a different lane;\n- a lane reporting clean over a surface another lane reported unreadable;\n- a split group whose sites landed in two lanes, where the `group` value is the only thing tying\n  them together.\n\nSo this pass runs even where every lane already verified itself, unless a `light` or `off` verification level skips it (Phase 5). It records `verified`, `inline`, or\n`skipped` **per lane** in the `verification` section, which is what keeps a lane that could not\nverify itself distinguishable from one that did.\n\n**Tell the verifier to flag only gaps affecting correctness or a stated requirement.** A reviewer\nasked for gaps will find some, and a verification pass that returns stylistic observations on every\nrun adds noise the operator has to re-triage each time while burying the class of gap it exists to\ncatch. The instruction is part of the dispatch, not an afterthought in the report.\n\n### Report\n\n**Two artifacts, `findings.json` and `report.md`**, written side by side in the run directory and\nredirected together by `--report-to \u003Cdir>`. The run prints `report.md`'s headline inline and offers\nthe file; it does not print the report, which is the cost the second artifact exists to avoid.\nBefore the headline, the inline message names what waits on the operator: the pass's human gate,\nthe `\u002Fdoctor` handoff when Phase 4 produced one, and any lane that stopped fenced or unverified.\n\nEvery run also emits, in **one line**, how many `OPINION`-tier checks were\navailable, how many were not run, and the argument that enables them (`--opinion`). Without it the\ntier ships unreachable.\n\n**Overlap notes are emitted here and never merge the findings they name.** Where two catalogs\nindependently flag the same site, the note names **both** `finding_id`s and leaves both findings\nstanding. Merging them would cross a plugin boundary the identity tuple keeps explicit: `check` is\nplugin-qualified, the two lanes persist independently, and an operator suppressing a merged finding\nwould be suppressing a catalog they never saw judged. The note sits in its own **note tier**,\nexcluded from the derived identity set and from P1-P3, because its inputs are judged findings held\nonly to P4's tolerance and a legitimate judged wobble must not fail a determinism property for\nbehavior P4 already licenses.\n\n## The suppression record\n\nA deliberately-kept finding is recorded in the `suppressions` mapping of\n`docs\u002Fconventions\u002Fharness-config.yaml` in the target repository (the convention home it binds, when\nit binds another), layered per the config-cascade convention. **It is the only suppression mechanism, and there is no inline\nmarker**, at any target: a marker would have to carry the same constituents, could not express a\ntwo-site finding at all, and would write into a tree the pass must leave clean. An entry stores the\nfinding's **constituents**, meaning `check`, `claim`, and every `(surface, anchor)` site, under the\n`finding_id` derived from them, never a bare id: an id is a one-way hash, so a record built on one\ncannot compute a tiered match. Keys, layer merge, precedence inversion, and the four entry\ndispositions are in [reference\u002Fsuppression.md](reference\u002Fsuppression.md); the cross-consumer key\ncontract is this marketplace's separately published **finding-suppression** convention.\n\n**Only the team layer enacts a suppression.** A personal entry for an id the team layer does not\ncarry is reported as `personal-only, not applied` rather than applied, since absence from the team layer\nis the team's *unsuppressed* state, so honoring a personal-only entry would hide a finding the team\nnever accepted. A suppression is a decision about the repository, so it belongs in the layer the\nrepository tracks; a personal layer drafts one for promotion.\n\nTwo report obligations. Every entry names its reason, its date, and **which cascade layer supplied\nit**. And **only an exact match is silent**: a one-sided anchor change carries forward as\n`needs-reconfirmation`, a deeper change closes the old entry and opens the new finding, and a\nvanished finding must be accounted for as a fix, a successor, a **retirement with its check** when\nthe check that raised it is absent or renamed in this run's detection configuration, or an\n**unexplained disappearance that fails the self-check**, the only detector the convergence property\nhas. Retirement is a reported disposition, not an exemption: it names the retiring check and the\nversion transition, because letting findings vanish silently on a catalog edit is the exact shape\nthis accounting exists to detect.\n\n## Self-check\n\n\u003C!-- fresh-eyes-exempt: deterministic-gate -- the tolerance comparison is set arithmetic over two runs' identity sets; its pass\u002Ffail IS the verdict and no judgment enters it -->\n**Establish the precondition first.** If HEAD or the state digest moved between the **scan-baseline**\nand **audit-endpoint** captures, or if two lanes recorded different content for a path they share, the\ntree did not hold still and the gate reports **`indeterminate`**, never `passed`. A shared checkout\nis the normal case, and an unfalsifiable pass manufactures confidence out of a basis nobody measured.\n\n**The audit endpoint is taken before Phase 5, not after it.** Under `--fix`, Phase 5 edits project\nfiles *by design*, so an endpoint captured after it necessarily differs from Phase 0, which would\nmark **every successful mutating run** `indeterminate` and skip P1–P3, the gate firing hardest on\nruns that did exactly what was asked. What the precondition is about is whether the tree held still\n*while the lanes were reading it*, and that window closes when the last lane completes. So the\ncapture bounds the read window, Phase 0 to end-of-lanes; Phase 5's writes fall outside the measured\ninterval rather than needing to be subtracted from it, which also avoids having to tell an accepted\nmutation apart from a coincidental identical one.\n\nA **third** capture is taken after Phase 5 and compared against the audit endpoint, for a different\npurpose: it confirms the applied set matches the accepted set, so a `--fix` run that changed\nsomething nobody approved is visible. That is a mutation-integrity check, reported separately from\nthe determinism one.\n\nWhen it held, any derived-tier inequality is a defect, and judged-tier growth beyond the tolerance in\n[reference\u002Fdeterminism-tiers.md](reference\u002Fdeterminism-tiers.md) **fails the self-check as an\ninstability finding against this skill**, naming the checks that moved, never absorbed by\nrecalibration.\n\n## Spoke paths\n\nThe `reference\u002F` files write this skill's directory as `\u003Cskill-dir>`, which is\n`${CLAUDE_SKILL_DIR}`, and the plugin's root directory as `\u003Cplugin-root>`, which is\n`${CLAUDE_PLUGIN_ROOT}`. Put those paths in place of the placeholders before running a command or\nwriting one into a brief. Those files arrive through the Read tool as plain bytes, so a `${…}` token\nin them would reach the Bash tool unsubstituted, and the Bash tool's environment has no\n`CLAUDE_SKILL_DIR` or `CLAUDE_PLUGIN_ROOT` to expand it from. Basis: the plugins reference,\n\u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fplugins\u002Fmanifest-reference#where-each-variable-resolves>.\n\n## Gotchas\n\n- **A suppression naming a registered cluster copy is a hard error, not a warning.** The copy is\n  excluded from the scan set, so an entry against it is stale by construction and the finding it\n  claims to hold belongs to the canonical source. Refuse; name that source.\n- **A whole-surface (`s:`) suppression survives every edit to the file and dies on a rename.** Not a\n  bug: a finding about a file *as a whole* must not be retired by editing a line inside it.\n- **A judged finding does not contribute to the determinism gate, which is the norm rather than a weakness.** Every\n  delegated check passes through model refinement first; the gate belongs to the derived tier.\n- **Never propose an `@path` import as a context saving.** A split remediation must name a\n  load-deferring destination and price what it costs. Pointer: when\n  judging whether a split saves context, fetch\n  [memory: import additional files](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmemory#import-additional-files)\n  and [memory: my CLAUDE.md is too large](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmemory#my-claude-md-is-too-large)\n  live.\n\n## What this skill does NOT do\n\n- Never defines a check. Adding criteria here rather than to the owning plugin's catalog is the\n  defect this skill's whole shape exists to avoid.\n- Never reads another plugin's files. Invocation-only cooperation, with one declared exception:\n  `retirements.yaml` is a published data file, read by the sweep lane via this plugin's own helper.\n- Never edits managed policy or a user-scope file, in any mode.\n- Never scans what it wrote. Where either resolved report path, `findings.json` or `report.md`, is\n  contained in the target, by `--report-to` or by `${CLAUDE_PLUGIN_DATA}` resolving under `~` for a\n  target at or above it, **both** paths are excluded before the write and the containment is\n  disclosed. Never silently, and never one of the two.\n- Never audits a target that is not a git repository. It refuses, and says what the refusal costs.\n","## Purpose\n\nOne bounded, ordered pass over a target repository's Claude Code **instruction** surface, coordinated\nacross three scopes and resumable mid-run. It **adds no criteria of its own**: every check is\ndelegated to the plugin that owns it. What it contributes is what invoking those skills by hand\nyields none of: a three-scope inventory before any check runs, a run-time-derived exclusion set,\nstable finding identity, suppression memory, incremental persistence and resume, and one human gate\nfor the pass. All of it is specified in [reference\u002Frun-contract.md](reference\u002Frun-contract.md).\n\n## Reference index. Load on demand\n\n| File | Load when |\n|---|---|\n| [reference\u002Fterms.md](reference\u002Fterms.md) | Before any other contract file. Every one of them uses run, target, lane, scan set, live surface and live surface set without redefining them. |\n| [reference\u002Ffinding-identity.md](reference\u002Ffinding-identity.md) | Emitting, comparing, or suppressing a finding: the `(check, claim, sites)` tuple, `surface`, `anchor`, and the derived `finding_id`. |\n| [reference\u002Frun-contract.md](reference\u002Frun-contract.md) | Deciding which contract file owns a rule, or resolving a `§N` cross-reference from inside one. |\n| [reference\u002Farguments.md](reference\u002Farguments.md) | A run is invoked with an argument whose precedence, default, or refusal is not settled by the `Arguments` table. |\n| [reference\u002Fexclusion-set.md](reference\u002Fexclusion-set.md) | Deriving the scan set, or justifying why a path was skipped. |\n| [reference\u002Frun-state-and-resumability.md](reference\u002Frun-state-and-resumability.md) | Starting, leasing, or resuming a run, and diagnosing a lock that outlived its holder. |\n| [reference\u002Freport-location-and-schema.md](reference\u002Freport-location-and-schema.md) | Writing the report, or judging what `--report-to` may target. |\n| [reference\u002Fdeterminism-tiers.md](reference\u002Fdeterminism-tiers.md) | Assigning a finding's tier, or running the self-check's comparison. |\n| [reference\u002Fsuppression.md](reference\u002Fsuppression.md) | Reading or writing the target's suppression record (`harness-config.yaml`). |\n| [reference\u002Fdoctor-handoff.md](reference\u002Fdoctor-handoff.md) | Reaching the `\u002Fdoctor` handoff, whether it is present or absent. |\n| [reference\u002Fliveness-sources.md](reference\u002Fliveness-sources.md) | Inventorying Phase 1 liveness: which sources this run can reach, what each one establishes, and the three losses it names in `skipped`. |\n| [reference\u002Flane-catalog.md](reference\u002Flane-catalog.md) | Dispatching Phase 3, or resolving what a `--lanes` id covers: every lane, what each delegated skill owns, what is not dispatched, and the substrate probe. |\n| [reference\u002Fretired-conventions-sweep.md](reference\u002Fretired-conventions-sweep.md) | Running the one script lane: how installed manifests are discovered, what an undiscoverable manifest set reports, and how a helper exit maps to a finding. |\n\n## Read-only contract, and where mutation can reach\n\nBare invocation reads and reports. `--fix` is the only mutation path, and it is bounded by scope:\n\n| Scope | Posture under `--fix` |\n|---|---|\n| Managed policy | **Never remediated.** Read-only in every mode. A finding here reports \"conflicts with org policy at `\u003Cpath>`\" and proposes no edit to either side; seeking a policy exception is an organizational decision, not a linting one. |\n| User | **Routed as a recommendation, never edited in place.** A user-scope tree is commonly managed by a dotfiles manager, so an in-place edit is drift the operator's own sync path will fight. |\n| Project | The only editable scope, per-finding confirmed. |\n\nThe frontmatter narrows the surface: `disallowed-tools: Edit, NotebookEdit` takes both editing\ntools away for a span the skills page bounds, so the read-and-report contract is instruction-held\nwith a narrowed accident surface, never an enforced guarantee. `Write` is kept, since run state and\nthe report persist under `${CLAUDE_PLUGIN_DATA}`, and it can mutate a file, so never describe the\nrestriction to an operator as mechanical. `audit-instructions` states the same posture.\n\n- **Pointer**: when deciding how long the restriction lasts, fetch [skills: frontmatter reference](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fskills#frontmatter-reference) live.\n\n**`scripts\u002Frun-state.sh` writes under that same plugin data directory and nowhere else**, never\ninside a target repository. It takes the data directory as an argument rather than discovering one,\nand validates both path segments it contributes: `lib\u002Fstate-key.sh` refuses a remote URL that would\nbecome traversing directory components, and a `--run-id` outside `[A-Za-z0-9][A-Za-z0-9_.-]*` is\nrefused here.\n\n**The run mechanics are scripts, not prose to re-derive per run.** Four of them, each with tests\nbeside it, and each is the *only* correct source for what it computes: a mechanic described in prose\nand performed by the model is a different mechanic per run, which is precisely what makes a\ndeterminism property unfalsifiable.\n\n| Script | Owns |\n|---|---|\n| `scripts\u002Frun-state.sh` | The run directory, the lease and its verdict, and the append-only partial. |\n| `scripts\u002Fstate-digest.sh` | `state-digest\u002Fv1` and `input-digest\u002Fv1`: the comparability input and every lane's resume digest. |\n| `scripts\u002Ffinding-identity.sh` | `anchor\u002Fv1` normalization, `finding_id\u002Fv1`, `group\u002Fv1`, and `guarded-append`: the guard, on the append path rather than beside it, so no record reaches the partial without passing it. |\n| `scripts\u002Fassemble.sh` | Partial-to-report assembly, and the `report.md` rendering of the assembled document. |\n\nThey read and write nothing inside the target: the digest scripts hash files the run names and print\na hex string, and the rest write only under the run directory.\n\n## Scope boundary (route out)\n\n- **One instruction surface against the model-capability catalog** → `\u002Fharness-config:audit-instructions`\n  directly. This pass dispatches that skill by invoking it via the Skill tool; it does not\n  re-answer it.\n- **Config-file correctness** is a delegated lane, not a route-out: this pass invokes\n  `\u002Fharness-config:audit` and appends the findings rows it persists (Phase 3).\n- **Grant portability** is a delegated lane at its `frontmatter` scope only (Phase 3). Its\n  settings-file scope routes out to `\u002Fharness-config:audit-permission-grants` directly, because the\n  permission plane as it is in effect is `audit-permission-state`'s lane here.\n- **Automation landscape** → `\u002Fharness-config:audit-automation-gaps`. Not in this pass's surface set:\n  it proposes automation that does not exist, so it reports over no inventoried surface.\n\n## Arguments\n\nParse `$ARGUMENTS`. The names and their one-line meanings:\n\n| Argument | Meaning |\n|---|---|\n| `target` | The git repository to audit. Defaults to the active project root; a `target` resolving anywhere else is refused non-zero. |\n| `--fix` | The explicit mutation override. Absent, the pass writes nothing into the target. |\n| `--opinion` | Run the `OPINION`-tier checks the delegated catalogs declare default-off. |\n| `--postures` | Dispatch the opt-in posture lane. Off by default: it is the additive catalog and the largest fan-out in the pass. |\n| `--lanes \u003Clist>` | Run only the named lanes, comma-separated. Narrows dispatch, never the inventory; an unknown lane id is refused non-zero. |\n| `--resume` | Resume the most recent incomplete run for this target's state key. |\n| `--report-to \u003Cdir>` | Redirect **both** report artifacts into that directory, subject to the destination gate. A directory, never a file path. |\n\nFlag semantics, precedence between them, and what `--fix` and `--resume` change about the phases\nbelow are in [reference\u002Farguments.md](reference\u002Farguments.md). Read it when a run is invoked with\nmore than one flag, or with any flag whose effect on a later phase you are about to assume.\n\n## Phase 0: Resolve, key, lock\n\nResolve the target root, compute the state key, and take the lock posture for the mode. Read-only\nruns take no lock and run concurrently; an applying run takes an exclusive advisory lock and refuses\nrather than queues. All specified in\n[reference\u002Frun-state-and-resumability.md](reference\u002Frun-state-and-resumability.md). With `--resume`,\nread the lease, then read the partial's lane records and carry forward every lane whose input digest\nis unchanged.\n\n**Do not derive the run directory or hand-write the lease.** `scripts\u002Frun-state.sh` does both, which\nis what makes this phase a mechanism rather than a description of one:\n\n```bash\nS=\"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Faudit-pass\u002Fscripts\u002Frun-state.sh\"\nD=\"${CLAUDE_PLUGIN_DATA}\"\nbash \"$S\" paths --plugin-data \"$D\" --run-id \"\u003Crun-id>\"\nbash \"$S\" lease acquire --run-dir \"\u003Crun-dir>\" --run-id \"\u003Crun-id>\" --plugin-data \"$D\"\n```\n\n`paths` derives `\u003Cplugin-data>\u002Fruns\u002F\u003Cstate-key>\u002F\u003Crun-id>` through the plugin's own `lib\u002Fstate-key.sh`,\nthe library whose header records the keying scheme. Pass `--plugin-data` explicitly with the `${CLAUDE_PLUGIN_DATA}` value from this text; never rely on the shell to expand it (Pointer: fetch [manifest reference: where each variable resolves](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fplugins\u002Fmanifest-reference#where-each-variable-resolves) live).\n`acquire` takes it too, and refuses a `--run-dir` that is not under\n`\u003Cplugin-data>\u002Fruns\u002F`. It is the only command that *creates* a directory, so it is where the write\ntree is pinned; a wrong or invented run dir would otherwise be created and written into.\n\n**`--resume` never attaches to a run that is still going.** Concurrent read-only runs are safe\nbecause each owns its own partial artifact; resume is the one operation that reaches into *another*\nrun's artifact, so the no-lock policy that makes concurrency safe is exactly what leaves resume\nunable to tell a live run from an interrupted one. Both would then append lane attempts and\nterminating records to one file, and highest-terminated-attempt assembly becomes race-dependent,\nthe interruption-tolerance mechanism producing a report neither run performed.\n\nSo every active run, read-only included, maintains a **lease**, and `--resume` reads it before it\nreads the partial. `run-state.sh lease classify --run-dir \u003Crun-dir>` prints the verdict:\na **live** lease means the run is still going, and resume exits non-zero naming the run id rather\nthan attaching; a **stale** lease means the run was interrupted and its artifact is resumable; a\n`released` tombstone is resumable immediately; `missing` means there is nothing to attach to. The\nlease is not a lock: it excludes nothing, blocks no concurrent read-only run, and grants no\nexclusivity; it answers the one question resume has to ask.\n\nRefresh it at every lane's persistence point (`lease heartbeat`) and write the tombstone on a clean\nexit (`lease release`). The full specification, covering path, contents, the two-sided liveness window, and an\nexplicit statement of **which clauses the script enforces and which remain the run's own discipline**,\nis in [reference\u002Frun-state-and-resumability.md](reference\u002Frun-state-and-resumability.md) §3. Read\nthat split before relying on any of it: a clause the script does not enforce is the run's own\ndiscipline, not a guarantee.\n\n**The scan baseline is captured after the inventory is frozen and before any lane reads.** The\ndigest spans every inventoried scope, so it cannot be computed before Phase 1 has produced that\ninventory. Taking it at the top of Phase 0 would either omit the user and managed surfaces, which\nare exactly the mid-run external edits the gate exists to detect, or force an unspecified second\ninventory. So Phase 0 resolves, keys, and locks; Phase 1 freezes the inventory; the **scan baseline**,\nthe target's HEAD commit and the run's state digest, is taken at that boundary, and the matching\n**audit endpoint** capture is taken when the last lane completes, before any Phase 5 mutation.\n\nBaseline to endpoint is therefore exactly the window in which lanes read, which is what the\ndeterminism gate is a claim about. A run that never measures it cannot claim it held. The digest\npairs each path with a hash of its current content, because a *count* holds still while a dirty\nfile's contents change underneath the run.\n\n## Phase 1: Three-scope inventory, before any check\n\n**Nothing is checked until all three scopes are inventoried.** A project-only inventory cannot see a\nproject-versus-user conflict, so a fix from one would be applied against half the picture.\nNative-first: the filesystem walk produces a **candidate** set, never the answer.\n\n**Liveness sources are the ones this run can actually reach, and a source it cannot reach is named\nrather than required.** Which sources this run can hold, which are out of reach and why, what each\none establishes, and the three coverage losses named in `skipped` rather than absorbed, are in\n[reference\u002Fliveness-sources.md](reference\u002Fliveness-sources.md). Read it before recording any\nliveness basis.\n\n**A liveness claim carries its basis, and the report marks a single-sourced one.** A single-sourced\ninventory is usable; silently presenting it as a multi-source result is the under-coverage-reads-as-\nclean failure the rule exists to prevent. The liveness basis records which sources were live, so a\nrun with an `InstructionsLoaded` producer and a run without are **not comparable** and cannot fail\nP1 against each other.\n\nRecord, per scope, every surface found **and every surface skipped with its reason**. The inventory\nis a reported derived-tier artifact, so a surface that silently drops out of scope between two runs\nfails the determinism property rather than looking like an improvement.\n\n**Output styles are the case a walk alone cannot get right.** A style changes the instructions\nClaude receives, and the live style can differ from the one the user's settings name, so a walk\nplus settings can report a selection as live when it is not. Report the resolved live style and\nwhat resolved it.\n\n- **Pointer**: when resolving the live style and what it changes, fetch [output styles: how output styles work](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Foutput-styles#how-output-styles-work) and [create a custom output style](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Foutput-styles#create-a-custom-output-style) live.\n\n**Shadowed definitions fall out of this inventory, not out of a catalog.** Where skills, subagents\nor MCP servers share a name across scopes, the pass compares names across the precedence order\nread live, deterministic, model-free and derived tier, and reports at `info` in its own section\nnaming the live definition and the shadowed one, never as a conflict finding or merged into one.\nWhen deciding which same-named definition is live, fetch [skills: resolve skills that share a name](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fskills#resolve-skills-that-share-a-name), [subagents: choose the subagent scope](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fsub-agents#choose-the-subagent-scope) and [MCP: scope hierarchy and precedence](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmcp#scope-hierarchy-and-precedence) live.\n\n## Phase 2: Derive the exclusion set\n\n**Derived at run time from the target's own state. Never transcribed, and no count of any class is\never carried in this skill.** A count written down is wrong on the next commit and wrong in every\nother repository. Four classes: registered byte-identical cluster copies, vendored upstream\nmaterializations, worktrees, and the pass's own artifacts. Each class's derivation, its fallback on a\nfailed read, and the hard error when a suppression targets an excluded path are in\n[reference\u002Fexclusion-set.md](reference\u002Fexclusion-set.md).\n\n## Phase 3: Lanes and dispatch\n\n**A lane is one delegated invocation at the finest filter that skill's own interface accepts**, the\ngranularity the pass can actually dispatch, since it invokes skills and never reaches inside one. A\nfiner lane is unbuildable, not merely inconvenient: per-lane persistence, input digests, and\nselective resume all key on something the pass can re-invoke on its own.\n\nSo the split falls out of the delegated interfaces rather than being asserted over them: a skill\ntaking a **surface-class filter** yields one lane per filter value it is given, and a skill taking\nonly an action verb yields **one lane** covering everything it audits. Where a skill runs its whole\ncatalog per invocation, the lane carries that whole catalog; the pass never splits a catalog it\ncannot address. Extending a delegated interface to accept a finer filter is a change to that skill,\nand until it lands the lane stays at the coarser grain.\n\n**What earns a lane, stated as a criterion rather than a list.** A skill is dispatched when it\nproduces findings **over surfaces the Phase 1 inventory already enumerates**, keyed to a site this\npass can anchor. That is what makes its rows joinable with the inventory, suppressible under this\npass's identity, and comparable across runs. `harness-config:audit-automation-gaps` fails the\ncriterion in both halves and is routed out: it is forward-looking, proposing automation that does not\nexist, so it has no inventoried surface to report over and no site to anchor a finding at.\n\n**Lane ids are what `--lanes` names**, one per lane in\n[reference\u002Flane-catalog.md](reference\u002Flane-catalog.md), plus one per surface-class value for the\ninstruction catalog: `instructions:\u003Cclass>`, `conflicts`, `permission-state`, `permission-grants`,\n`config`, `memory`, `retirements`, and `postures`. A `--lanes` value naming an id outside that set is\nrefused non-zero.\n\nDispatch, in inventory order, with every skill invoked via the Skill tool, each invocation\npresence-gated with its fallback stated. Every lane, what each delegated skill owns, why each is\nexactly one lane (or one per surface-class value), what is deliberately left undispatched, and the\nsubstrate probe that decides whether a lane can spawn its own verifier, are in\n[reference\u002Flane-catalog.md](reference\u002Flane-catalog.md). Load it before dispatching a lane, before\nanswering what a `--lanes` id covers, and before assuming a lane can verify itself.\n\nPersist each lane's findings to the partial artifact **as that lane completes**, never buffered to\nthe end. A lane is complete when its terminating record is in the partial, and every record carries\nits attempt id so an abandoned re-attempt is discardable rather than merely older. The write is one\ncall per record, and it goes through the guard rather than past it:\n\n```bash\nFI=\"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Faudit-pass\u002Fscripts\u002Ffinding-identity.sh\"\nbash \"$FI\" guarded-append --run-dir \"\u003Crun-dir>\" --record '\u003Cjson-line>' --epoch \"\u003Cheld>\"\n```\n\n`guarded-append` runs `validate-record` and appends only if it passes, so **no record reaches the\npartial without clearing §1**; `run-state.sh partial append` is what it calls and is never the call a\nlane makes itself. **Exit 4 is the guard's refusal and nothing was written**: a null identity, three\nor more sites, two sites with no pairwise declaration, an anchor outside the grammar, or a\n`finding_id\u002Fv1` disagreeing with its constituents. Fix the lane, never the record. The lease is\nrefreshed at the same boundary; a lease must exist, so a record resume could not attribute to a\nlive-or-abandoned run is never written. **Pass the epoch you hold**: the filename is the writer's\nepoch, not whatever the lease now carries, which keeps a fenced writer's rows out of its adopter's\nfile. The record is also validated as well-formed single-line JSON rather than sniffed by its first\ncharacter, because a malformed row here is permanent and resume is its only reader.\n\n**Exit 3 means FENCED: stop this run.** The record was written safely to your own epoch file, but the\nlease has moved on and another run has adopted the artifact, so continuing dispatches lanes whose\noutput nothing will assemble. Stop and report the run as superseded. Never retry the append, and never\nread that exit as transient.\n\n**The lane count is bounded by the delegated interfaces, not chosen here**, at one per scope value the\ninstruction catalog accepts plus one each for the permission plane, grants, config, the memory layer,\nthe retirements sweep, and the opt-in posture lane, so it is a handful, and a per-run\ndispatch ceiling would never bind. What is *not* bounded here is the fan-out inside a lane: the\ndelegated catalogs spawn their own subagents, and this pass cannot reach inside one to cap it. So cap\nconcurrency at 3–5 lanes and let incremental persistence carry the rest. It is what degrades a blown\nsession ceiling into a resumed run.\n\nThe guarded-append call above bounds nothing inside a lane; what it buys is that an intra-lane\noverrun costs the lanes still running rather than the whole pass.\n\n## Phase 4: The `\u002Fdoctor` handoff\n\nThis pass builds no replacement for `\u002Fdoctor` and hands the instruction-file audit to `\u002Fdoctor\nprompt-audit` rather than running it; for what each owns, fetch [memory: audit your instruction files](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmemory#audit-your-instruction-files)\nlive. **Both are\noperator-run, so neither is dispatched.** Their records (version floors, write posture),\nwhat the presence check verifies versus probes, and the absence classification are in\n[reference\u002Fdoctor-handoff.md](reference\u002Fdoctor-handoff.md).\nWhen `\u002Fdoctor` is absent, name it as the missing capability and state what goes unchecked.\n\n**Phase 4 records the handoff; it does not stop the pass.** Halting here would mean a `--fix` run\nnever reaches Phase 5 and no run reaches the Phase 6 report, so an optional collaborator would\ncancel the coordinated pass that is this skill's purpose. `reference\u002Fdoctor-handoff.md` says to\nfinish the pass's own phases first.\n\nSo Phase 4 opens the `delegated` lane, records the instruction in the report, and continues. Phases 5\nand 6 run normally, and the assembled report carries the handoff as an outstanding item with its lane\nmarked `open`, routed and not yet returned. The operator runs `\u002Fdoctor` when they choose; a later\n`--resume` closes that lane by re-prompt rather than re-scan, because nothing about it needs the\nsweep to run again.\n\n**`open` is an assembly terminator, not a completion.** The two are distinct and conflating them\nwould have made the promised resume impossible: §7 needs a terminating record to assemble a report at\nall, while §5 skips any lane whose state is complete and whose digest is unchanged, so a lane that\nwas both terminated *and* complete would be carried forward untouched on every resume, and the\noutstanding handoff would never close. So the record terminates the attempt for assembly and marks\nthe lane's state **incomplete**. `--resume` therefore re-runs it, which for a delegated lane means\nre-prompting rather than re-scanning. `handed-back` and `declined` are completions; only `open` is\nnot.\n\n**That instruction to the operator is only true if the terminating record is actually written.**\n`--resume` reads the partial, not the report, so a report telling the operator to come back with\n`--resume` against a partial nothing wrote is a false instruction in the one artifact they act on. So\nthe `open` terminator is written at the moment Phase 4 records the handoff, never deferred to Phase 6\nassembly, which is exactly where a run that does not reach Phase 6 loses it. It goes through\n`guarded-append` like every other record: the guard short-circuits on anything that is not a\n`finding`, so guarding a terminator costs nothing, and **one write command for the whole skill** is\nwhat keeps Phase 3's rule from reading as advice a coordinator may opt out of.\n\n## Phase 5: Apply, only under `--fix`\n\nPer-finding confirmation, project scope only, bounded by the scope table under \"Read-only contract, and where mutation can reach\". Refuse and name the\ncanonical source if a fix or a suppression would write into a derived-exclusion path.\n\nThe apply-verify step judges work this same run produced, so it is delegated: hand the applied diff\nand the finding it claims to resolve, the artifact rather than this run's reasoning, to a cross-vendor\nadvisor when one is installed and set up (the OpenAI Codex plugin, say, invoked per its own docs),\nwith a **fresh-context (non-fork) subagent** as the stated fallback. Resolve the `verification` level once per run with `bash \"${CLAUDE_PLUGIN_ROOT}\u002Flib\u002Fread-verification.sh\" --plugin harness-config --skill audit-pass` and copy its line into the report header ([levels](https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fdocs\u002Fcontracts\u002Fverification-level\u002FREADME.md)). Under `light`, and `off`, which acts as `light` here, neither this apply-verify dispatch nor Phase 6's main-session verification runs and nothing is verified inline in their place: record `skipped` for the reason `verification: skipped (\u003Clevel>)` and mark findings and fixes `(unverified)`, as \"When dispatch is unavailable\" below says. A delegated lane applies its own skill's level.\n\n### When dispatch is unavailable\n\nThe apply-verify step and delegated lanes that mandate subagent dispatch **require** that dispatch.\nWhen the Agent tool is blocked, unavailable, or the probed depth budget does not permit the spawn a\nstep needs:\n\n1. **Record per-lane verification mode** in the lane's terminating record and the assembled report (`verified` |\n   `inline` | `skipped`) for every lane that mandates independent verification.\n2. **Mark unverified findings.** Proposals or applied fixes that did not receive an independent\n   verifier carry an `(unverified)` marker and are never presented as resolved.\n3. **Do not silently complete.** The `skipped` section and report header name dispatch\n   unavailability, and the probed depth budget, when either prevented a mandated verification phase.\n4. **The main-session pass is the one that survives a lowered depth budget**, because it spawns from\n   the top of the session rather than from inside a lane. A run that could verify no lane inline can\n   usually still verify the assembled set, so `inline` per lane and a completed Phase 6 verification\n   is the ordinary shape on a depth-1 host, not a degraded one.\n\n## Phase 6: Assemble, verify the assembled set, report\n\n**Assemble first.** An append-only `findings.partial.\u003Cowner_epoch>.jsonl` during the run becomes\n`findings.json` here, epoch-scoped so a fenced writer cannot interleave into its adopter's file, with\nassembly reading only the highest epoch present and, per lane, the highest-ordinal attempt carrying a\nterminating record. `scripts\u002Fassemble.sh` performs it. Both schemas, the identity-versus-presentation\nfield split, and the sections are in\n[reference\u002Freport-location-and-schema.md](reference\u002Freport-location-and-schema.md); the tier\nmemberships are in [reference\u002Fdeterminism-tiers.md](reference\u002Fdeterminism-tiers.md). Tiers stay in\nseparate sections because their guarantees differ.\n\n### Main-session verification, after assembly\n\n**A verification pass runs here, in the main session, on the assembled set, and its justification is\nscope rather than inability.** A lane can only ever verify its own findings; several relations exist\nonly once every lane's output is in one document, and nothing inside a lane can see them:\n\n- a finding two catalogs both raised at one site, which is the overlap note's whole subject;\n- a finding contradicting a suppression the record carries for a different lane;\n- a lane reporting clean over a surface another lane reported unreadable;\n- a split group whose sites landed in two lanes, where the `group` value is the only thing tying\n  them together.\n\nSo this pass runs even where every lane already verified itself, unless a `light` or `off` verification level skips it (Phase 5). It records `verified`, `inline`, or\n`skipped` **per lane** in the `verification` section, which is what keeps a lane that could not\nverify itself distinguishable from one that did.\n\n**Tell the verifier to flag only gaps affecting correctness or a stated requirement.** A reviewer\nasked for gaps will find some, and a verification pass that returns stylistic observations on every\nrun adds noise the operator has to re-triage each time while burying the class of gap it exists to\ncatch. The instruction is part of the dispatch, not an afterthought in the report.\n\n### Report\n\n**Two artifacts, `findings.json` and `report.md`**, written side by side in the run directory and\nredirected together by `--report-to \u003Cdir>`. The run prints `report.md`'s headline inline and offers\nthe file; it does not print the report, which is the cost the second artifact exists to avoid.\nBefore the headline, the inline message names what waits on the operator: the pass's human gate,\nthe `\u002Fdoctor` handoff when Phase 4 produced one, and any lane that stopped fenced or unverified.\n\nEvery run also emits, in **one line**, how many `OPINION`-tier checks were\navailable, how many were not run, and the argument that enables them (`--opinion`). Without it the\ntier ships unreachable.\n\n**Overlap notes are emitted here and never merge the findings they name.** Where two catalogs\nindependently flag the same site, the note names **both** `finding_id`s and leaves both findings\nstanding. Merging them would cross a plugin boundary the identity tuple keeps explicit: `check` is\nplugin-qualified, the two lanes persist independently, and an operator suppressing a merged finding\nwould be suppressing a catalog they never saw judged. The note sits in its own **note tier**,\nexcluded from the derived identity set and from P1-P3, because its inputs are judged findings held\nonly to P4's tolerance and a legitimate judged wobble must not fail a determinism property for\nbehavior P4 already licenses.\n\n## The suppression record\n\nA deliberately-kept finding is recorded in the `suppressions` mapping of\n`docs\u002Fconventions\u002Fharness-config.yaml` in the target repository (the convention home it binds, when\nit binds another), layered per the config-cascade convention. **It is the only suppression mechanism, and there is no inline\nmarker**, at any target: a marker would have to carry the same constituents, could not express a\ntwo-site finding at all, and would write into a tree the pass must leave clean. An entry stores the\nfinding's **constituents**, meaning `check`, `claim`, and every `(surface, anchor)` site, under the\n`finding_id` derived from them, never a bare id: an id is a one-way hash, so a record built on one\ncannot compute a tiered match. Keys, layer merge, precedence inversion, and the four entry\ndispositions are in [reference\u002Fsuppression.md](reference\u002Fsuppression.md); the cross-consumer key\ncontract is this marketplace's separately published **finding-suppression** convention.\n\n**Only the team layer enacts a suppression.** A personal entry for an id the team layer does not\ncarry is reported as `personal-only, not applied` rather than applied, since absence from the team layer\nis the team's *unsuppressed* state, so honoring a personal-only entry would hide a finding the team\nnever accepted. A suppression is a decision about the repository, so it belongs in the layer the\nrepository tracks; a personal layer drafts one for promotion.\n\nTwo report obligations. Every entry names its reason, its date, and **which cascade layer supplied\nit**. And **only an exact match is silent**: a one-sided anchor change carries forward as\n`needs-reconfirmation`, a deeper change closes the old entry and opens the new finding, and a\nvanished finding must be accounted for as a fix, a successor, a **retirement with its check** when\nthe check that raised it is absent or renamed in this run's detection configuration, or an\n**unexplained disappearance that fails the self-check**, the only detector the convergence property\nhas. Retirement is a reported disposition, not an exemption: it names the retiring check and the\nversion transition, because letting findings vanish silently on a catalog edit is the exact shape\nthis accounting exists to detect.\n\n## Self-check\n\n\u003C!-- fresh-eyes-exempt: deterministic-gate -- the tolerance comparison is set arithmetic over two runs' identity sets; its pass\u002Ffail IS the verdict and no judgment enters it -->\n**Establish the precondition first.** If HEAD or the state digest moved between the **scan-baseline**\nand **audit-endpoint** captures, or if two lanes recorded different content for a path they share, the\ntree did not hold still and the gate reports **`indeterminate`**, never `passed`. A shared checkout\nis the normal case, and an unfalsifiable pass manufactures confidence out of a basis nobody measured.\n\n**The audit endpoint is taken before Phase 5, not after it.** Under `--fix`, Phase 5 edits project\nfiles *by design*, so an endpoint captured after it necessarily differs from Phase 0, which would\nmark **every successful mutating run** `indeterminate` and skip P1–P3, the gate firing hardest on\nruns that did exactly what was asked. What the precondition is about is whether the tree held still\n*while the lanes were reading it*, and that window closes when the last lane completes. So the\ncapture bounds the read window, Phase 0 to end-of-lanes; Phase 5's writes fall outside the measured\ninterval rather than needing to be subtracted from it, which also avoids having to tell an accepted\nmutation apart from a coincidental identical one.\n\nA **third** capture is taken after Phase 5 and compared against the audit endpoint, for a different\npurpose: it confirms the applied set matches the accepted set, so a `--fix` run that changed\nsomething nobody approved is visible. That is a mutation-integrity check, reported separately from\nthe determinism one.\n\nWhen it held, any derived-tier inequality is a defect, and judged-tier growth beyond the tolerance in\n[reference\u002Fdeterminism-tiers.md](reference\u002Fdeterminism-tiers.md) **fails the self-check as an\ninstability finding against this skill**, naming the checks that moved, never absorbed by\nrecalibration.\n\n## Spoke paths\n\nThe `reference\u002F` files write this skill's directory as `\u003Cskill-dir>`, which is\n`${CLAUDE_SKILL_DIR}`, and the plugin's root directory as `\u003Cplugin-root>`, which is\n`${CLAUDE_PLUGIN_ROOT}`. Put those paths in place of the placeholders before running a command or\nwriting one into a brief. Those files arrive through the Read tool as plain bytes, so a `${…}` token\nin them would reach the Bash tool unsubstituted, and the Bash tool's environment has no\n`CLAUDE_SKILL_DIR` or `CLAUDE_PLUGIN_ROOT` to expand it from. Basis: the plugins reference,\n\u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fplugins\u002Fmanifest-reference#where-each-variable-resolves>.\n\n## Gotchas\n\n- **A suppression naming a registered cluster copy is a hard error, not a warning.** The copy is\n  excluded from the scan set, so an entry against it is stale by construction and the finding it\n  claims to hold belongs to the canonical source. Refuse; name that source.\n- **A whole-surface (`s:`) suppression survives every edit to the file and dies on a rename.** Not a\n  bug: a finding about a file *as a whole* must not be retired by editing a line inside it.\n- **A judged finding does not contribute to the determinism gate, which is the norm rather than a weakness.** Every\n  delegated check passes through model refinement first; the gate belongs to the derived tier.\n- **Never propose an `@path` import as a context saving.** A split remediation must name a\n  load-deferring destination and price what it costs. Pointer: when\n  judging whether a split saves context, fetch\n  [memory: import additional files](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmemory#import-additional-files)\n  and [memory: my CLAUDE.md is too large](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmemory#my-claude-md-is-too-large)\n  live.\n\n## What this skill does NOT do\n\n- Never defines a check. Adding criteria here rather than to the owning plugin's catalog is the\n  defect this skill's whole shape exists to avoid.\n- Never reads another plugin's files. Invocation-only cooperation, with one declared exception:\n  `retirements.yaml` is a published data file, read by the sweep lane via this plugin's own helper.\n- Never edits managed policy or a user-scope file, in any mode.\n- Never scans what it wrote. Where either resolved report path, `findings.json` or `report.md`, is\n  contained in the target, by `--report-to` or by `${CLAUDE_PLUGIN_DATA}` resolving under `~` for a\n  target at or above it, **both** paths are excluded before the write and the containment is\n  disclosed. Never silently, and never one of the two.\n- Never audits a target that is not a git repository. It refuses, and says what the refusal costs.\n",{"metadata":24,"description":7,"argument-hint":27,"user-invocable":28,"disallowed-tools":29,"disable-model-invocation":20},{"summary":25,"workflow-stage":26},"Run one coordinated, resumable audit pass over a repo with a single human gate","anytime","[target] [--fix] [--opinion] [--postures] [--lanes \u003Clist>] [--resume] [--report-to \u003Cdir>]",true,"Edit, NotebookEdit","main","d0e68c3f36b92a1d6381f0ba0060d7a0b8b886c5","plugins\u002Fharness-config\u002Fskills\u002Faudit-pass",7,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins","https:\u002F\u002Fgithub.com\u002Fmelodic-software","https:\u002F\u002Fmelodicsoftware.com","Independent software studio building developer tools, automation, and application services, with roots in music.","2026-11-15 05:15:06",{"source":8,"skillName":5,"commands":40,"prompt":50,"zipAvailable":28},[41,44,47],{"tool":42,"command":43},"npx","npx skills add melodic-software\u002Fclaude-code-plugins --skill audit-pass",{"tool":45,"command":46},"bunx","bunx skills add melodic-software\u002Fclaude-code-plugins --skill audit-pass",{"tool":48,"command":49},"pnpm","pnpm dlx skills add melodic-software\u002Fclaude-code-plugins --skill audit-pass","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002FSKILL.md",[52,58,62,66,70,74,78,82,86,90,94,98,102,106,110,116,120,124,128,132,136,140],{"path":53,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":56,"sourceHtmlUrl":57},"evals\u002Fevals.json","REFERENCE","参考文档",41437,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fevals\u002Fevals.json",{"path":59,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":60,"sourceHtmlUrl":61},"reference\u002Farguments.md",10876,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Farguments.md",{"path":63,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":64,"sourceHtmlUrl":65},"reference\u002Fdeterminism-tiers.md",26350,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Fdeterminism-tiers.md",{"path":67,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":68,"sourceHtmlUrl":69},"reference\u002Fdoctor-handoff.md",6992,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Fdoctor-handoff.md",{"path":71,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":72,"sourceHtmlUrl":73},"reference\u002Fexclusion-set.md",6351,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Fexclusion-set.md",{"path":75,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":76,"sourceHtmlUrl":77},"reference\u002Ffinding-identity.md",21382,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Ffinding-identity.md",{"path":79,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":80,"sourceHtmlUrl":81},"reference\u002Flane-catalog.md",9858,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Flane-catalog.md",{"path":83,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":84,"sourceHtmlUrl":85},"reference\u002Fliveness-sources.md",6262,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Fliveness-sources.md",{"path":87,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":88,"sourceHtmlUrl":89},"reference\u002Freport-location-and-schema.md",19379,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Freport-location-and-schema.md",{"path":91,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":92,"sourceHtmlUrl":93},"reference\u002Fretired-conventions-sweep.md",7918,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Fretired-conventions-sweep.md",{"path":95,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":96,"sourceHtmlUrl":97},"reference\u002Frun-contract.md",1341,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Frun-contract.md",{"path":99,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":100,"sourceHtmlUrl":101},"reference\u002Frun-state-and-resumability.md",28117,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Frun-state-and-resumability.md",{"path":103,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":104,"sourceHtmlUrl":105},"reference\u002Fsuppression.md",13727,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Fsuppression.md",{"path":107,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":108,"sourceHtmlUrl":109},"reference\u002Fterms.md",1234,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Freference\u002Fterms.md",{"path":111,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":114,"sourceHtmlUrl":115},"scripts\u002Fassemble.sh","SCRIPT","脚本",17044,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fscripts\u002Fassemble.sh",{"path":117,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":118,"sourceHtmlUrl":119},"scripts\u002Fassemble.test.sh",16242,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fscripts\u002Fassemble.test.sh",{"path":121,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":122,"sourceHtmlUrl":123},"scripts\u002Ffinding-identity.sh",19870,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fscripts\u002Ffinding-identity.sh",{"path":125,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":126,"sourceHtmlUrl":127},"scripts\u002Ffinding-identity.test.sh",23396,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fscripts\u002Ffinding-identity.test.sh",{"path":129,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":130,"sourceHtmlUrl":131},"scripts\u002Frun-state.sh",29906,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fscripts\u002Frun-state.sh",{"path":133,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":134,"sourceHtmlUrl":135},"scripts\u002Frun-state.test.sh",24205,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fscripts\u002Frun-state.test.sh",{"path":137,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":138,"sourceHtmlUrl":139},"scripts\u002Fstate-digest.sh",14658,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fscripts\u002Fstate-digest.sh",{"path":141,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":142,"sourceHtmlUrl":143},"scripts\u002Fstate-digest.test.sh",12916,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fharness-config\u002Fskills\u002Faudit-pass\u002Fscripts\u002Fstate-digest.test.sh",[145,152,159,166,171,178],{"skillNo":146,"name":147,"nameZh":6,"description":148,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":149,"categories":150,"keywords":151,"publishedAt":19},"01M7DYYDG31B9YX7N9MB9QW2GW","track","Backlog reads through the bound tracker: search open and closed items, view one item with its comments and state history. Use when another skill searches or reads tracker items.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fevals\u002Frecall-still-reports-reopened-bug\u002Fwork-items-stub\u002Fskills\u002Ftrack\u002FSKILL.md",[],[],{"skillNo":153,"name":154,"nameZh":6,"description":155,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":156,"categories":157,"keywords":158,"publishedAt":19},"01M7DYXVBAVG3NV30PA7GKW9Q6","reviewer-media","Reviewer media for one pull request: decides whether it needs a narrated walkthrough, builds the storyboard from \u002Freview:explain-change's digest, has \u002Fexplainer-video:produce render it and hands the manifest to \u002Fsource-control:pull-request media. Use when asked for 'reviewer media', 'a walkthrough video for this PR', 'a produced video of this change', or when pull-request ready or monitor hands a PR over.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Freview\u002Fskills\u002Freviewer-media\u002FSKILL.md",[],[],{"skillNo":160,"name":161,"nameZh":6,"description":162,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":163,"categories":164,"keywords":165,"publishedAt":19},"01M7DYRZ5YD44AMYH26YR25GZ8","route","Model and effort per agent role (orchestrator, worker, verifier, retrieval) for a multi-agent run, resolved through the bundled defaults and the user, team and personal override config layers into fenced JSON with single-agent and fan-out variants and the layer behind each value. Use when asked 'which model for the subagents', 'route the roles' or 'model routing', or before a skill passes a role's model on a spawn.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fdiscovery\u002Fevals\u002Fex-model-pin-without-override\u002Fmulti-agent-stub\u002Fskills\u002Froute\u002FSKILL.md",[],[],{"skillNo":167,"name":161,"nameZh":6,"description":162,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":168,"categories":169,"keywords":170,"publishedAt":19},"01M7DYRTVT4FWTGDJY0HCCFJMZ","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fdiscovery\u002Fevals\u002Fex-model-override-from-route\u002Fmulti-agent-stub\u002Fskills\u002Froute\u002FSKILL.md",[],[],{"skillNo":172,"name":173,"nameZh":6,"description":174,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":175,"categories":176,"keywords":177,"publishedAt":19},"01M7DYNZC8PNW3T27V2ADKECKJ","principles","Module-design vocabulary for architecture findings, and the checks built on it. Use when: 'what is a deep module', 'what is a seam', 'deep vs shallow module', 'what is the deletion test', 'leverage and locality', 'module design vocabulary', 'which word do we use for X in architecture findings'. Scanning a codebase for friction is \u002Farchitecture:improve.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Farchitecture\u002Fskills\u002Fprinciples\u002FSKILL.md",[],[],{"skillNo":179,"name":180,"nameZh":6,"description":181,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":182,"categories":183,"keywords":184,"publishedAt":19},"01M7DYNMCZNEWBFHV9KTSS6H60","write","Structured five-field bug report (title, steps to reproduce, expected vs actual, justified severity, suggested fix location) from an informal description, read-only. Use when the user names a defect they observed, or asks for one written up ('report a bug', 'file a bug', 'write this up as a bug').","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Farchitecture\u002Fevals\u002Fdeepening-defect-goes-to-bugs-write\u002Fbugs-stub\u002Fskills\u002Fwrite\u002FSKILL.md",[],[]]