fix(claude-ops): repair inventory for 2.1.284 and discover native overlaps dynamically

docs/conventions/native-references/CHANGELOG.md

@@ -6,6 +6,15 @@ major change; additive guidance is minor; clarification is a patch. The doc ship
 unnumbered, which this file reads as **1.0**; the entry below is the first recorded change and lands
 the changelog the README said would arrive with it.
 
+## [3.2.0] - 2026-09-29
+
+Minor: additive guidance.
+
+- **`bundled-workflow` is a provenance class.** Claude Code bundles workflows (`deep-research`)
+  beside its skills, and the overlap store can now record a row against one. Its runtime
+  relationship is `route` or `suggest`, never `wrap`, because the Native step invokes through the
+  Skill tool and a workflow is not a skill.
+
 ## [3.1.0] - 2026-09-29
 
 Minor: additive guidance in the Runtime relationship section.

docs/conventions/native-references/README.md

@@ -236,6 +236,7 @@ at runtime. Every store row carries one of `route`, `wrap`, or `suggest`.
 | `builtin-command` | `route` or `suggest`. Never `wrap`: a built-in command is not invoked by the Skill tool |
 | `bundled-skill` | `route` or `wrap` |
 | `bundled-skill` carrying `model-invocation-disabled` | `suggest` only. The model never lists the surface, so a route phrase is dead text. The marker is set from the registration the row's evidence names, never from the bare name |
+| `bundled-workflow` | `route` or `suggest`. Never `wrap`: the Native step invokes through the Skill tool, and a workflow is not a skill |
 | `plugin-backed-builtin` | `route` or `wrap` |
 | `marketplace-plugin` | `route` or `wrap`. The wrap grammar for this class is seam-phrasing's, not the Native step below |
 | `session-skill` | `route` only |

docs/native-surfaces.md

@@ -19,6 +19,7 @@ and when. See [`docs/conventions/native-references/`](conventions/native-referen
 |---|---|---|---|---|
 | Built-in CLI commands | 5 | 5 | route 1, suggest 4 | complementary 5 |
 | Bundled skills | 13 | 12 | route 8, suggest 3, wrap 2 | complementary 12, defer 1 |
+| Bundled workflows | 0 | 0 | none | none |
 | Plugin-backed built-ins | 1 | 1 | route 1 | complementary 1 |
 | Session-provided skills (observation-only) | 1 | 0 | route 1 | defer 1 |
 | First-party marketplace plugins | 2 | 2 | route 2 | complementary 2 |
@@ -335,6 +336,10 @@ and when. See [`docs/conventions/native-references/`](conventions/native-referen
 - **Baked:** description phrase yes · Boundary section yes · Native step no · suggest sentence no
 - **Budget caveat:** the baked phrase may be dropped from the skill listing under budget pressure. It is the best available routing surface, not a guaranteed one
 
+## Bundled workflows
+
+No rows recorded in this lane.
+
 ## Plugin-backed built-ins
 
 ### `security-review` → `review:security-review`

plugins/claude-ops/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
   "name": "claude-ops",
-  "version": "0.65.0",
+  "version": "0.66.0",
   "description": "Claude Code operations toolkit. Thirteen skills: audit-skill-visibility (audit whether each installed skill is actually VISIBLE to the model, and diagnose why most of a fleet never gets used: a skill is invisible when its description is dropped by Claude Code's skill-listing context budget, which sheds descriptions lowest-score-first so an unused skill loses the keywords that would let it be matched, from skills genuinely not wanted, from skills the run cannot observe at all; computes whether the listing overflows from documented settings, and withholds every cold verdict the data cannot support rather than reporting absence of data as absence of use), inventory (read-only enumeration of the complete invocable surface: every built-in CLI command with aliases and hidden/gated status, every bundled skill, and every component of every installed plugin across all marketplaces; reads the shipped binary because upstream publishes no built-in command list, and carries an integrity verdict so a drifted build reports counts as floors rather than silently short totals), audit-install-state (read-only audit of the machine-scope ~/.claude installation directory and ~/.claude.json: full inventory split into an authored surface and rolled-up bulk trees, product-managed retention vs genuinely unmanaged state, filename-scheme resolution before any process-liveness check, and deliberate/mid-experiment detection; reports, never deletes), audit-performance (read-only slowness-diagnostic capture run at the moment the machine or a session feels slow: CLI version, retention-sweep health including the unparsable-settings pause, which warns in /status, a timed census walk of the install tree as a sweep-cost proxy, active-session and plugin-fleet counts, a process census, and the fan-out layer, which covers a load-labeled no-op spawn baseline, every hook that will fire bucketed per-tool-call versus per-turn with its invocation shape, the configured statusline, subagent concurrency and spawn-depth ceilings against documented defaults, whether running sessions predate the settings file they are judged by, and orphan attribution by parent liveness rather than age, plus on Windows a kernel-object census (Token objects against uptime, paged pool) that names a host-level leak beneath all four suspects; read against a bundled known-performance-issues reference that also records the causes tested and cleared; separates the four documented suspects of accumulated state, version regression, component bloat, and per-spawn fan-out cost, and routes remediation out; reports, never mutates, and never executes a discovered hook or statusline command), audit-native-overlap (map native Claude Code surfaces, namely built-in CLI commands, bundled skills, plugin-backed built-ins, and session-provided skills, against the current repo's plugin skills and agents, so a custom component never silently duplicates what Claude Code itself ships; bare invocation is a read-only overlap report carrying the extraction's integrity floors and a shared-listing-budget exposure section, verdicts are human-gated in a committed store rendered into a generated registry whose every row carries an observable recheck trigger, and only an explicit apply step bakes presence-gated native references into descriptions and Boundary sections), observability (read locally captured telemetry from the OTEL store, the collector, the per-session hook event log and hook-event JSONL, and ccusage, with trend reports, a per-session report of what fired, what was blocked and the event timeline, and store pruning), known-issues (search known Claude product GitHub bugs, check service health, maintain a persistent tracked-issue registry), changelog (ingest Claude Code changelog entries and integrate them into the current repo), prerequisites (read-only table of external binaries declared by enabled plugins; never installs), plugins (bring a machine's plugin fleet current on demand: marketplace refresh, effective-scope updates including in-repo project/local installs, new-plugin install per policy, scope-divergence detection and explicit convergence), morning-brief (read-only gh-based operator morning view: queue-label counts, merge-ready PRs, parked decisions with their RECOMMENDED lines, and loop-lane telemetry freshness), lanes (start/restart/stop/status loop lanes as named background Claude Code sessions seeded from canonical prompt files, with per-lane model/effort, a repo-pull + marketplace-refresh launch step, and a consume-restarts action, an OS-schedulable reader that relaunches stopped lanes whose telemetry carries a restart_request), and a re-runnable setup action that settles where the known-issues registry, the skill-usage log and the hook log root live, places the root's self-ignoring guard, and detects retired conventions. Plus an opt-in, default-off per-session hook event log (one JSON line per hook event on every event the generated registry marks observable, written to <root>/sessions/<session_id>.jsonl, with SessionEnd retention by session count or age and an optional detached pre-prune command), a family of eight advisory *-audit hooks (API errors, config changes, instruction loads, permission denials, pre-compaction, skill usage, tool failures, and unsurfaced hook failures. The last also warns the user via systemMessage, since a hook that fails to launch enforces nothing and Claude Code surfaces the failure to nobody) that emit the shared hook-telemetry envelope, and a reference sink that routes envelopes under the same root: per session when the envelope carries a session id, else into the shared hook-events.jsonl the observability skill reads.",
   "author": {
     "name": "Melodic Software",

plugins/claude-ops/CHANGELOG.md

@@ -3,6 +3,41 @@
 All notable changes to the `claude-ops` plugin are documented here. Format follows
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); this plugin uses semantic versioning.
 
+## [0.66.0] - 2026-09-29
+
+### Added
+
+- **`inventory` reports every built-in surface's arguments and invocability.** Commands, bundled
+  skills and bundled workflows carry `argument_hint`, `user_invocable` and `model_invocable`,
+  resolved from the bundle; a value the bundle computes at runtime is `null` and counted under
+  `integrity.undetermined`, never guessed.
+- **`inventory` has a bundled-workflows lane** (`bundled_workflows`, canary `deep-research`) with
+  its own integrity status.
+- **`inventory --docs`** fetches the live commands page and changelog and classifies each name:
+  documented, undocumented, alias, removed in the docs, removed but still registered, or
+  docs-only, with kind and alias disagreements. A fetch failure degrades that block only.
+  `--docs-file` and `--changelog-file` run it offline.
+- **`audit-native-overlap detect` discovers candidates** by scoring every native surface against
+  every repo skill and agent (`--threshold`, `--top-k`), beside the seeded pairs. Each candidate
+  carries `invocable_by` and a `recommended_integration` label: a user-only surface is
+  recommended as `suggest`. A label is never a verdict.
+
+### Fixed
+
+- **`inventory` extraction on Claude Code 2.1.284.** Template-literal substitutions are now read as
+  code, so a quote inside a regex in `${...}` no longer desynchronizes the brace reader (15 of 152
+  commands resolved before). `registerSlidesSkill`, literal-table skill rosters and
+  constant-named commands resolve. Validated against 2.1.284.
+- **`inventory --docs` bounds untrusted text.** A fetched body over 16 MB degrades the docs block
+  instead of loading, and a table row over 8,000 characters is skipped, so a malformed page
+  cannot stall the parser on regex backtracking. A body truncated after its headers
+  (`http.client.HTTPException`) degrades the block instead of raising.
+- **`audit-native-overlap detect` scores a plugin-backed command once**, under its plugin-backed
+  class and with the description the extractor enriched it with, instead of adding a bare second
+  surface.
+- **A model-invocable bundled workflow is recommended `route`, never `wrap`**, matching the store
+  rule that rejects `wrap` on a bundled-workflow row.
+
 ## [0.65.0] - 2026-09-29
 
 ### Added

plugins/claude-ops/skills/audit-native-overlap/SKILL.md

@@ -84,7 +84,10 @@ python3 "${CLAUDE_PLUGIN_ROOT}/skills/audit-native-overlap/scripts/overlap.py" s
 ```
 
 `--repo`, `--store`, `--view`, and `--pairs` are flags with repo-relative defaults, so a consumer
-repository with a different layout points them wherever its files live.
+repository with a different layout points them wherever its files live. `detect` also takes
+`--threshold` (lowest discovery score kept, default `0.30`) and `--top-k` (most components kept per
+native surface, default `3`; `0` turns discovery off). Lower the threshold to `0.25` for a
+recall-first sweep; below that, most added pairs share one incidental word.
 
 Every subcommand exits `0` ok, `1` broken, `3` degraded, the sibling extractor's contract, not the
 shell gates' `0/1/2`, so one lane can carry both; `2` stays argparse's usage error. A degraded exit
@@ -94,10 +97,35 @@ for the repo's test discovery.
 
 ## Detection posture. Floor-honest
 
-Under-recall stated honestly beats confident completeness. Three rules:
+Under-recall stated honestly beats confident completeness. Candidates come from two origins:
+
+- **Seeded**: every pair in `reference/canonical-pairs.json`, emitted whether or not the extraction
+  shows the native side.
+- **Discovered**: every native surface in the extraction, internal ones excepted, scored against
+  every skill and agent in the repo from name, alias, and description tokens
+  (`scripts/discover.py` states the formula). A pair at or over `--threshold`, among the `--top-k`
+  best for its surface, is emitted with its score and the shared tokens as evidence. A pair the
+  store already records is listed under `discovery.existing` with its verdict instead; a pair that
+  is also seeded stays one seeded candidate carrying the score.
+
+Discovery is lexical. A one-line native description rarely shares words with the component it
+duplicates in concept (`recap` against `session-flow:orient`), so such a pair belongs in the seeds
+once a human confirms it; a pair discovery already finds needs no seed.
+
+Every candidate carries `native.invocable_by` (`model+user`, `user-only`, `model-only`, or
+`unknown`), read from the registration's `model_invocable` and `user_invocable` fields, or from
+`disable_model_invocation` in an older extraction; a field the extraction lacks makes it `unknown`.
+`model_invocable: false` also sets the `model-invocation-disabled` marker the store's suggest-only
+rule reads. From it comes `recommended_integration`: `suggest` for a user-only surface, `route` or
+`route-or-wrap` for a model-invocable one (`route` for a built-in command or a bundled workflow,
+which never take `wrap`), and nothing when unknown. It is a label for the human writing the row, never a store value.
+
+Three rules:
 
 - **Carry the integrity floor through, per lane.** The inventory reports integrity per lane
-  (`builtin_commands`, `bundled_skills`, `plugin_backed`). A `degraded` lane makes every count from
+  (`builtin_commands`, `bundled_skills`, `plugin_backed`, and `bundled_workflows` when the
+  extraction has that lane; an extraction without it is not an error, the lane is simply not
+  reported). A `degraded` lane makes every count from
   that lane a floor, and the report says so in the same sentence as the number. A `broken` lane's
   counts are omitted, the report names the lane and its cause, and every candidate whose lane is
   broken is marked `re_derivable: false` (its presence or absence in that lane proves nothing
@@ -118,9 +146,11 @@ Inventory status per lane (ok | degraded | broken), cli_version vs validated_aga
 that means for every count below; a broken lane is named with its cause.
 
 ## Overlap candidates
-One row per (native surface, our component): native name + provenance class + hidden/gated
-markers, our component, the evidence, and the store's current verdict — or NEW where the
-store has no row yet.
+One row per (native surface, our component): origin (seeded | discovered), native name +
+provenance class + hidden/gated markers, invocable_by, our component, score and shared tokens,
+recommended integration (a label, not a verdict), the evidence, and the store's current verdict,
+or NEW where the store has no row yet. Discovered pairs the store already records follow as one
+line each with their verdict.
 
 ## Registry state
 Rows whose recheck trigger has fired, rows missing a baked line, rows baked but unverified.
@@ -133,8 +163,8 @@ to name-only degradation.
 Which substrate produced which section, and anything the run could not resolve.
 ```
 
-Provenance classes are never merged into one list. A bundled skill, a built-in command, a
-plugin-backed built-in, and a session-provided skill have different disable switches and different
+Provenance classes are never merged into one list. A bundled skill, a bundled workflow, a built-in
+command, a plugin-backed built-in, and a session-provided skill have different disable switches and different
 rosters per host; a merged list cannot be acted on.
 
 ## Budget exposure, a presence-gated seam
@@ -308,14 +338,15 @@ Two upstream facts this skill depends on, each with the trigger that obliges re-
 - **A plugin skill never shadows a native one.** Ours are namespaced, so both resolve and the model
   chooses. That is why the routing lives in descriptions rather than in a name.
 - **`plugin_backed` is its own lane.** `security-review` is reported there, not under
-  `builtin_commands`. Read the wrong key and the row looks absent. Verified 2026-09-06 against
-  Claude Code 2.1.263, by running `inventory.py --binary-only` on this machine and reading the
-  `plugin_backed` key, which holds `security-review` and nothing else. Recheck when the extractor's
-  provenance lanes change or a release note moves a bundled surface between them.
+  `builtin_commands`. Read the wrong key and the row looks absent. Verified 2026-09-29 against
+  Claude Code 2.1.284, by reading the `plugin_backed` key of an `inventory.py --binary-only`
+  extraction on this machine, which holds `security-review` and nothing else. Recheck when the
+  extractor's provenance lanes change or a release note moves a bundled surface between them.
 - **A bundled skill can carry aliases.** `code-review` answers to `review`; treating an alias as a
   separate surface produces a duplicate row for one capability. Basis:
   <https://code.claude.com/docs/en/commands> gives `/code-review` the line "Alias: `/review`".
-  Verified 2026-09-06 against Claude Code 2.1.263 and that page as fetched that day. Recheck when
+  Verified 2026-09-29 against Claude Code 2.1.284 (the extraction lists `review` as the alias) and
+  that page as fetched that day. Recheck when
   the commands page drops the alias line or a release note renames a bundled skill.
 - **Absent from the binary is not absent from the product.** Session-provided skills exist only in
   a live roster. "Not in the extraction" is a statement about the extraction.