Attribution audit: restated upstream facts left after the upstream-pointer sweep, and a check-stamps trigger-form gap

@TSV

Description

After the upstream-pointer sweep, /attribution:audit ran report-only over every markdown file the sweep's branches changed, in three groups (A, B and C, 28 branches in all). It found no fingerprint-confirmed copies in any group, so nothing is fix-eligible. It did find restated upstream facts: a default, limit or behavior owned by an external source, stated as current with no whole record (anchor, as-of date, recheck trigger) beside it. Every finding below was unanimous across three blind judges and survived a default-to-refute adversary. All are report-only.

Line numbers are the nominators' approximate ranges on the sweep branches, read 2026-10-11. Confirm each line on main before editing.

Counts by tier

group	restated-fact	tier split	copy findings	split panels sent to a human
A	138	138 `llm-suspected` (panels batched; marked DEGRADED)	0	42
B	220	187 `source-fetched-similar`, 33 `not-found`	0	79
C	167	95 `source-fetched-similar`, 72 `llm-suspected`	3	82
total	525		3	203

llm-suspected means no source was fetched for the fact; source-fetched-similar means a fetched page backs it; not-found means the source was not fetched within the budget, and concludes nothing about the passage.

The three copy findings (paraphrase or near-verbatim, review survived, report-only):

plugins/playbooks/reference/model-adaptation/fable-5-1.md:26-28 (source: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1#batch-independent-tool-calls-in-agent-loops)
plugins/playbooks/reference/model-adaptation/fable-5-1.md:37-41 (source: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1#ask-for-user-facing-progress-updates)
plugins/session-flow/skills/workflow/context/steps.md:155-156 (source: https://code.claude.com/docs/en/best-practices#add-an-adversarial-review-step)

Top recurring classes

Taken from the reports' own cross-branch readings, not a separate tally:

The generated "How to set these" README block. The largest class. scripts/sync-plugin-options-docs.py writes it into every plugin README that declares options (50 on main). It states Claude Code CLI and settings behavior (the -s default scope, already installed still writing the value, uninstall dropping pluginConfigs, which settings files plugin options are read from) and defers its record to the plugin-reconfiguration convention. At least 23 findings across the groups are this block, several standing for every README in a branch. Fixing the template once in scripts/sync-plugin-options-docs.py covers the class: Regenerate the plugin-options README blocks with a pointer-only template after the open PRs touching them merge #7291.
A relay to another file's dated record. A harness fact is stated with "the record lives in ", without carrying that record's date or trigger. Judges graded this as no record (see open question 1).
Third-party tool and API behavior with no source. gh and GitHub REST paging, search and closing-keyword behavior, git ref resolution, Windows and Microsoft tooling, formatter and linter CLI flags. Most not-found and llm-suspected rows are this class; the work-items GitHub adapter README alone carries 15.

Stamp-detector gap in check-stamps

plugins/attribution/skills/audit/scripts/check-stamps.sh:491 recognizes a recheck trigger only as recheck ... trigger, trigger for|is|: or what would reopen. It misses the inline form the sweep used everywhere, "(as of 2026-10-11; recheck when ...)", and the **Recheck:** ... form. Every rule-trigger-less-stamp flag in all three groups was this miss: group C flagged 57 stamps and 55 carried an inline "recheck when" clause; group B flagged 34 and all 34 did; group A's flags were all the same false positive.

Repro: write a file containing Verified 2026-10-01 (as of 2026-10-01; recheck when the hooks page renames this section). and run check-stamps.sh --trigger-less <file>. The stamp is reported as rule-trigger-less-stamp (reproduced on main, 2026-10-12).

The detector also under-fires by design (check-stamps.sh:27-32, :487): one trigger phrase anywhere in a file clears every stamp in it, so "trigger-less 0" is not a clean result.

Open rubric questions

These readings move many verdicts; decide them before acting on the list.

Does a record kept in a sibling file count toward R4? Judges and refuters graded R4 against the containing file only, so a passage that says "the dated record lives in sources.md" was treated as having no record. Several dozen findings rest on that reading and would clear if a cross-file record counts.
Does a file-level default stamp plus a bare pointer make a whole record? For example, a header that gives every pointer in the file an as-of date and a "docs change touching the section it points at" trigger. Judges split on this; it accounts for most of the harness-memory splits.

Separately, .claude/rules/skill-bodies-state-current-rules.md ("A record never cures a restatement") is stricter than the audit rubric. The audit applied the rubric only.

Findings by plugin

Each row: file, approximate lines, and the restated fact in brief.

actionlint (5)

plugins/actionlint/README.md
- 31-36: "ShellCheck deadlocks on large blocks under the …
- 37-42: "The transcript shows a hook error notice, and lint …
- 46-48 (same line in 5 other READMEs): "install Git for Windows so Claude Code can run it …
- 97-98: "actionlint auto-discovers its own …
- 130-189 (same generated block in all 9 READMEs): "-s defaults to user"; uninstall drops …

adhd (3)

plugins/adhd/skills/clarify/SKILL.md
- 208-212: It ships with the artifact surface, so it is normally …
plugins/adhd/README.md
- 90-93: Deterministic every-session arming needs a …
- 63-65: context compaction/summarization in a long session can …

architecture (4)

plugins/architecture/skills/map-containers/SKILL.md
- 296-299: "It refuses mermaid because mermaid C4 is …
plugins/architecture/skills/map-context/SKILL.md
- 226-229: "which refuses mermaid because mermaid C4 is …
plugins/architecture/skills/map-flow/SKILL.md
- 231-240: "as do the same quoted strings on MapGet/MapPost …
plugins/architecture/skills/map-landscape/SKILL.md
- 238-243: "Structurizr has a dedicated systemLandscape view …

autonomy (14)

plugins/autonomy/README.md
- 138-143: "/goal's session-only completion condition"; "a …
- 160-172: `--settings …
- 183-187: "That value reaches the hook only as the forgeable env …
- 273-289: generated options table: "Every option Claude Code …
- 293-315: "prints already installed and still writes the …
- 317-338: "Plugin option values are read from user …
plugins/autonomy/reference/prerequisite-resolution.md
- 71-78: enableAllProjectMcpServers / enabledMcpjsonServers …
plugins/autonomy/skills/setup/SKILL.md
- 30-34: "Resolve ${CLAUDE_PROJECT_DIR} (fall back to the …
- 116-125: "CLAUDE_CODE_ENABLE_TELEMETRY=1, per-signal …
plugins/autonomy/skills/setup/context/guardrail-slice.md
- 53-59: Windows drive mount "(/mnt/<letter>/)"
plugins/autonomy/skills/setup/context/trigger-dispatch-slice.md
- 16-22: "event-trigger workflows must exist on the default …
- 23-27: "shortest interval 5 minutes"; "public repos …
plugins/autonomy/skills/setup/templates/ci-otlp-artifact.md
- 57-82: "The agent CLI exports OTLP over the network only" …
- 86-100: CLAUDE_CODE_ENABLE_TELEMETRY=1 …

bash-format (3)

plugins/bash-format/README.md
- 23-34: "no parser/printer flags, so your .editorconfig is …
- 70-72: "ShellCheck discovers .shellcheckrc by walking up …
- 84-92: "Claude Code's own session scratchpad would otherwise …

biome-format (1)

plugins/biome-format/README.md
- 31-37: "biome check --write applies safe fixes, formatting …

bugs (13)

plugins/bugs/README.md
- 86-90: the plugin data directory is ephemeral on cloud and …
- 96-102: Claude Code prompts for userConfig at enable time
- 144-145: Claude Code prompts at enable time and /plugin …
- 150-160: reinstall writes config, -s defaults to user, and …
- 161-164: hook option env is fixed at session start
- 186-187: CLAUDE_PLUGIN_OPTION_* env vars carry configured …
- 207-214: gh issue create --type sets the native Issue Type, an …
plugins/bugs/skills/scan/SKILL.md
- 74-78: the plugin data directory is ephemeral on cloud …
plugins/bugs/skills/setup/SKILL.md
- 56-64: reinstall writes config, uninstall drops …
- 65-66: user_config is rendered at skill load
plugins/bugs/skills/write/SKILL.md
- 96: Claude Code's Read tool cannot read video files
- 117: CLAUDE_PLUGIN_DATA is scoped per plugin, not per …
- 126: gh --type sets the native Issue Type, org repos only

code-metrics (3)

plugins/code-metrics/skills/principles/SKILL.md
- 60-64: multimetric reports Halstead per file; radon per …
- 133-134: OMG's current ASCQM release is v1.1.
- 174-175: One of CRAP's two expansions currently appears on the …

code-tidying (13)

plugins/code-tidying/skills/audit-dead-code/SKILL.md
- 118-122: "includeDeclaration is hard-coded true, so the dead …
- 137-138: "knip exits 1 for findings and for hard errors …
- 203-205: "gopls check -severity=hint reports through every …
plugins/code-tidying/skills/audit-dead-code/context/adjudication.md
- 29-31: same LSP findReferences fact; L57-60 record covers …
- 66-76: "ignore", "ignoreExportsUsedInFile": true …
- 78-87: "a whitelist file passed alongside the sources"
plugins/code-tidying/skills/batch-simplify/SKILL.md
- 60-67: "the harness runs a skill's whole pre-compute block as …
- 90-94: "git's --since approxidate parser requires `24 …
plugins/code-tidying/skills/batch-simplify/context/repo-mode.md
- 217-232: "a single untracked path in the list aborts the …
plugins/code-tidying/skills/tidy/lanes/self-update.md
- 3: "Consumers receive updates through `/plugin …
- 51: "A broken --- fence in a SKILL.md would block the …
plugins/code-tidying/skills/tidy/reference/exclusions.md
- 21: "a plugin's hooks/hooks.json and a skill's or …
- 125: "userConfig types are string, number, boolean …

computer-use (7)

plugins/computer-use/README.md
- 22-27: Computer-use screenshots use a fixed pixel budget …
plugins/computer-use/skills/diagnose/SKILL.md
- 75-79: Computer-use screenshots land at a fixed megapixel …
plugins/computer-use/skills/setup/SKILL.md
- 44-46: list_granted_applications is read-only and returns the …
- 60-61: request_access raises a consent dialog.
- 65-66: Claude's synthesized input does not reset the OS idle …
- 106-107: With a focus-stealing window in front, the …
- 123-125: powercfg /requests requires elevation and exits …

context-budget (14)

plugins/context-budget/README.md
- 6-9: "/context report only as totals: the built-in tool …
- 111-120: "Claude Code's if file rules do not match an …
- 129-132: "listed skill frontmatter is subtracted from that …
- 185-187: generated options table: "Every option Claude Code …
- 197-212: "prints already installed and still writes the …
- 214-217: "Hooks are handed their CLAUDE_PLUGIN_OPTION_* when …
- 219-240: "Plugin option values are read from user …
plugins/context-budget/skills/audit/SKILL.md
- 15-24: same /context totals fact; L97-107 record's trigger …
- 55-57: "explains where the current session's tokens went …
- 69-70: "The skill is gated, and bundled skills vary by …
- 312-321: "Listed skill-frontmatter tokens are subtracted" …
plugins/context-budget/skills/audit/reference/engine.md
- 68-75: "The prefix System tools bucket has listed …
plugins/context-budget/skills/audit/reference/report.md
- 51-52: "The listing cap is characters, not the Skills token …
- 57-58: "/skill-doctor for unused skills and /doctor for …

context-guard (10)

plugins/context-guard/README.md
- 105-107: "Claude Code offers no way to withdraw a shown …
- 139-142: "an organization policy can refuse a user mod's tools"
- 146-148: "$.session.usage() carries no cache expiry" …
- 160: "A plugin cannot add a status-line segment"; status …
- 300: "Install at user scope (the default)"
- 381-383: generated options table: "Every option Claude Code …
- 399-444: generated How-to-set block: already installed, -s …
plugins/context-guard/reference/reader-contract.md
- 503-504: "Auto-compact offers no steering hook"
- 712-719: "${CLAUDE_PLUGIN_DATA} resolves …
plugins/context-guard/skills/check/SKILL.md
- 20: "A separate claude plugin test process never sees …

context7 (3)

plugins/context7/skills/lookup/context/mcp.md
- 54-59: mcp__context7__resolve-library-id ... …
plugins/context7/skills/lookup/SKILL.md
- 117-117: ctx7 docs /org/project "..." gets path-mangled to …
- 22-55: If mcp__context7__resolve-library-id / …

desktop-notification (2)

plugins/desktop-notification/README.md
- 132-166: "Plugin option values are read from user …
- 37-38: "Install libnotify-bin (Debian/Ubuntu) or …

discipline (5)

plugins/discipline/skills/setup/SKILL.md
- 18-23: "stores non-sensitive options in user settings, and …
- 39-43: "An option the user never configured does not reliably …
- 82-91: "prints already installed and still writes the …
- 92-94: "the rendered ${user_config.*} values are injected …
plugins/discipline/skills/sweep-all/SKILL.md
- 100-105: "An unset option does not reliably substitute to empty …

discovery (24)

plugins/discovery/agents/intent-tracer.md
- 18-21: "a failed preload is skipped silently in the harness …
- 41-46: "the target does not reach a preloaded body by …
- 86-91: "skipped silently: Claude Code logs a warning to …
- 103-106: "You hold every tool the harness offers a subagent ... …
- 113-118: "An allowlist removes all MCP tools"
- 220-224: maxTurns: "one assistant turn may hold several …
- 320-324: "a harness older than v2.1.246 may return none"
plugins/discovery/agents/researcher.md
- 18-22: same preload fact as intent-tracer.md:18-21
- 59-64: "the topic does not reach a preloaded body by argument …
- 83-91: "that pin is why a substituted effort in a preloaded …
- 115-120: silent skip, debug-log warning
- 132-144: inherited pool incl. every MCP tool; "An allowlist …
- 242-246: maxTurns counting unit
- 340-344: "a harness older than v2.1.246 may return none"
plugins/discovery/skills/explore/reference/workflow.md
- 30: root CLAUDE.md "already in context, in a dispatched …
plugins/discovery/skills/research/SKILL.md
- 118-122: "If that still reads as a literal placeholder ... the …
- 214: "The topic does not reach a preloaded body by argument …
- 225-230: "the harness runs a skill's whole pre-compute block as …
- 288-294: /deep-research "fans out web searches, fetches and …
- 308-310: "The workflow needs the WebSearch tool"
plugins/discovery/skills/research/context/discipline.md
- 281-288: "(50k URL / 50 MB cap per file)"; /llms.txt (or …
plugins/discovery/skills/trace-intent/SKILL.md
- 21-26: pre-compute block / worktree compound-command fact …
- 58-62: "a preload that fails to resolve is skipped …
plugins/discovery/skills/trace-intent/context/dispatch.md
- 107-110: "a subagent that reaches maxTurns returns its output …

disk-hygiene (12)

plugins/disk-hygiene/README.md
- 189-194: "rejecting the zero-length WindowsApps\python3.exe …
- 227-233: "disable the python3 App execution alias (Settings > …
- 563-576: "A skill-frontmatter hook receives only …
- 655-700: generated How-to-set block: -s defaults to user …
plugins/disk-hygiene/skills/clean/SKILL.md
- 69: "*.img, which includes WSL's ext4.vhdx"
plugins/disk-hygiene/skills/clean/reference/unsupported-platform-handoff.md
- 154-160: "bin size caps, a policy-disabled bin, or a …
- 161-166: "A path longer than the classic Windows MAX_PATH …
plugins/disk-hygiene/skills/setup/SKILL.md
- 86-90: "the skill-frontmatter belt that Claude Code keeps …
- 98-123: "a zero-length WindowsApps\python3.exe App Execution …
- 142-149: same Settings path; "bare python3 still opens the …
- 166-172: "managed settings or a --settings flag can carry a …
- 186-202: "prints already installed **and still writes the …

docs-hygiene (10)

plugins/docs-hygiene/skills/audit-progressive-disclosure/SKILL.md
- 24-29: "the harness runs a skill's whole pre-compute block as …
- 33-36: "invocation-loaded content recurs for the rest of the …
- 49-55: "CLAUDE.md above ~200"; "Reference file >300 lines …
- 162-174: "the 200-line CLAUDE.md target"; "two Anthropic …
plugins/docs-hygiene/skills/extract-ssot/actions/batch.md
- 159-169: "concurrent agents editing the same file silently …
plugins/docs-hygiene/skills/extract-ssot/context/anti-patterns.md
- 43: "Anthropic's 'head -100' partial-read failure mode …
- 129: "self-generated skills provide NO benefit on average" …
plugins/docs-hygiene/skills/rename-references/SKILL.md
- 27-32: same pre-compute / worktree fact; relay only
- 162-165: "Edit tool's read-before-write guard catches racing …
plugins/docs-hygiene/skills/write-for-agents/reference/agent-doc-surfaces.md
- 136-147: cross-vendor table: ".cursor/rules/*.mdc ... Always …

dometrain (12)

plugins/dometrain/skills/setup/SKILL.md
- 18-21: A plugin-provided server's name carries its plugin's …
- 56-60: A connection-layer 401/403/429 rejection (per …
- 64-71: Claude Code's native prompt collects the key ... …
- 80-88: needs /reload-plugins or a new session ... `claude …
- 147-153: Both plugins ship defaultEnabled: false, so they …
- 155-158: /plugin configure dometrain-mcp@<marketplace> …
- 167-171: Dometrain ships its own official Claude Code plugin …
plugins/dometrain/README.md
- 3-7: Dometrain's hosted server …
- 33-40: DOMETRAIN_API_KEY, the variable Dometrain's own …
- 73-86: Dometrain ships its own official Claude Code plugin …
- 94-97: That plugin's .mcp.json authenticates via a shell …
- 99-104: Install identity is marketplace-scoped ... skill and …

eol-normalizer (1)

plugins/eol-normalizer/README.md
- 20-28: "NUL scan of the first 8000 bytes, git's own detection …

evals (18)

plugins/evals/README.md
- 32-37: claude plugin eval runs each case with and without the …
- 44-47: The plugin eval CLI reads evals/ cases as prompt.md …
- 72-76: --max-cost-usd bounds one invocation, not a session …
- 111-112: --max-cost-usd bounds one invocation and a run that …
- 124-125: Claude Code prompts for plugin options on enable, and …
- 161-164: pluginConfigs option values are read from user …
- 166-167: Claude Code passes configured plugin option values to …
plugins/evals/skills/methodology/reference/eval-design.md
- 50-54: On models where thinking is always on, thinking tokens …
plugins/evals/skills/plugin-eval/reference/case-authoring.md
- 18-34: evals/ is the default eval dir (overridden by …
- 47-57: Each run gets a throwaway workspace; scaffold_script …
- 64-75: llm and baseline graders cost three judge calls per …
- 88-95: regex grader match modes (not_contains, count:N); the …
- 111-118: MCP tools are mocked from evals/mocks by default, a …
- 161-172: The regex grader takes flags: i, not inline (?i); an …
plugins/evals/skills/plugin-eval/reference/ci.md
- 81-93: An incomparable case omits delta; skippedPaidGraders …
- 99-104: Under --ablation none no grader is excluded from …
plugins/evals/skills/plugin-eval/reference/reading-results.md
- 3-4: With --json , claude plugin eval prints no …
- 112-114: Run score is the weighted fraction of graders passed …

event-storming (11)

plugins/event-storming/skills/simulation/reference/miro-integration.md
- 15-17: the hosted server exposes no board-delete tool (Miro's …
- 53-54: Get a token from …
- 68-87: ### Miro's 16 Sticky Note Colors ... light_yellow …
- 110-117: REST API supports geometry.width ... all stickies …
- 136-136: Miro uses x,y coordinates with (0,0) at canvas center. …
- 237-240: Setting parent_id switches x and y from …
- 294-298: The sharing policy must be nested under …
plugins/event-storming/skills/simulation/SKILL.md
- 41-50: Its tools resolve under the plugin-provided prefix …
- 61-63: The miro plugin must be installed from an available …
- 185-190: ${CLAUDE_PLUGIN_DATA}/ (the per-plugin directory …
- 22-31: Executes all 5 sub-rounds: Financial value -> …

fleet (6)

plugins/fleet/skills/reach/SKILL.md
- 102-104: notify_when_idle from a -p sender does not work …
- 208-209: claude -p reads stdin (Claude Code CLI behavior)
- 217-218: With --, the distro's login shell re-parses the rest …
- 221-222: SessionEnd hooks print Hook cancelled on stdout …
plugins/fleet/skills/reach/reference/relay.md
- 39-41: claude.exe needs a Windows-side working directory …
- 106-108: The CLI docs do not specify the stdin message schema …

go-format (3)

plugins/go-format/README.md
- 29-34: "goimports itself has no awareness of that …
- 43-50: -local grouping prefix from go list -m
- 69-79: "go install golang.org/x/tools/cmd/goimports@latest"

harness-config (55)

plugins/harness-config/reference/agents-md-liveness.md
- 36-40: A session that loads no instruction files at all …
plugins/harness-config/skills/audit-automation-gaps/context/hook-timing.md
- 41-46: PermissionRequest / Not via exit code 2 / Denial …
- 49-51: async: true is valid only on type: "command" …
- 58-63: Add an if rule so a non-matching call returns before …
plugins/harness-config/skills/audit-instructions/SKILL.md
- 475-477: Keep the 3–5 format/tone/structure-steering examples …
plugins/harness-config/skills/audit-instructions/context/phase-a-inventory.md
- 11-17: setting CLAUDE_CONFIG_DIR relocates the whole …
- 20-26: a file is natively read only where no CLAUDE.md …
- 129-140: CLAUDE_CODE_DISABLE_AUTO_MEMORY is authoritative …
- 197-200: the cache can hold several versions of one plugin, and …
plugins/harness-config/skills/audit-pass/SKILL.md
- 200-202: Skills, subagents, and MCP servers override by name …
- 288-292: /doctor owns the CLAUDE.md trim-and-migrate half …
plugins/harness-config/skills/audit-pass/reference/doctor-handoff.md
- 129-130: /doctor is a prompt-based bundled skill -- Claude …
plugins/harness-config/skills/audit-pass/reference/liveness-sources.md
- 75-76: They come from a decay-weighted usage score -- how …
plugins/harness-config/skills/audit-permission-grants/SKILL.md
- 155-156: /permissions (built-in command, alias …
- 100-102: "Always allow" writes there, and nothing prunes it. -- …
- 144-146: ${CLAUDE_PLUGIN_ROOT} and ${CLAUDE_PLUGIN_DATA} …
plugins/harness-config/skills/audit-permission-grants/reference/criteria.md
- 20-23: That scope is where Claude Code's own "Always allow" …
- 85-99: an action class Claude Code drops on entering auto …
- 101-113: Monitor allow rules are a further documented drop …
- 157-160: Bash(...) / A glob over the literal command string …
- 186-189: Bash rules match literally and do not expand …
- 247-249: a plugin settings.json supports only the agent and …
plugins/harness-config/skills/audit-permission-state/SKILL.md
- 15-20: there is no claude permissions subcommand or …
- 146-158: A user-level deny blocks a project-level allow just as …
- 205-208: **autoMode.classifyAllShell inverts the answer …
- 230-232: We treat only the string "disable" as locking …
- 275-283: evaluation still runs deny, then ask, then allow …
- 291-296: the server-managed cache at …
- 343-348: the reader resolves local **through worktrees to the …
plugins/harness-config/skills/audit-permission-state/reference/criteria.md
- 166-185: a deny rule that is a bare tool name as removing the …
- 326-338: C6-malformed: 'Text after the closing parenthesis, as …
- 371-376: a customized section omits "$defaults", which …
plugins/harness-config/skills/audit-permission-state/reference/gotchas.md
- 34-40: settings.local.json resolves through worktrees to …
- 44-47: the repository-root copy wins on a shared key but …
- 48-51: the operator's own user settings and local file are …
plugins/harness-config/skills/audit-prompting-postures/SKILL.md
- 82-97: under the claude-md-and-agents-md setting both files …
- 170-178: ${CLAUDE_PLUGIN_DATA} is machine-global, not …
plugins/harness-config/skills/audit/context/procedures.md
- 11-17: opens the file from inside a subprocess, which a …
- 19-22: the baseline Read deny merges into the sandbox …
- 60-69: In auto mode it needs the [Self-Modification] …
- 76-80: disableAllHooks, the managed allowManagedHooksOnly …
- 87-92: Deny and ask rules are evaluated regardless of what a …
plugins/harness-config/skills/audit/context/validation-categories.md
- 110-112: timeout is a seconds value ... the docs give …
- 279-286: The listing budget is a fraction of the model's …
- 294-304: /doctor estimates the listing's cost and its biggest …
- 336-341: The /skills menu saves there from a home-rooted …
- 391-396: The declared settings schema types the key `"type" …
plugins/harness-config/skills/audit/reference/audit-checklist.md
- 49-49: Bug [feat(planning): Claude proposes merges and admits that the host approves with one tap #8961]: deny rules in local are silently ignored …
plugins/harness-config/skills/audit/reference/native-update-config.md
- 25-27: Bundled skills can be removed by …
plugins/harness-config/skills/draft-auto-mode-rules/SKILL.md
- 49-51: /auto-mode-setup (built-in command): drafts …
- 65-66: The command is hidden, gated by plan and version, and …
- 152-154: The classifier reads autoMode from user settings …
- 165-166: claude auto-mode reset ... strips the autoMode …
plugins/harness-config/skills/unhobble/SKILL.md
- 197-202: inventory every component type that same table lists …
- 250-262: a session reads them as the project instructions only …

harness-memory (9)

plugins/harness-memory/skills/audit/SKILL.md
- 16-18: "The root-file line count has @ imports expanded …
- 201-207: "its claude-md-improver skill audits CLAUDE.md …
plugins/harness-memory/skills/stateless/SKILL.md
- 37-48: "<user-config-dir>/projects/<project>/memory/" …
- 90-107: "/memory (built-in command): the person's …
- 22-29: "one directory per repo …
- 60-67: "1 reports auto memory off, and 0 reports it …
plugins/harness-memory/skills/stateless/context/purge.md
- 28-39: "autoMemoryDirectory as able to relocate the store …
plugins/harness-memory/skills/stateless/context/status.md
- 38-48: "CLAUDE_CODE_DISABLE_AUTO_MEMORY overrides …
plugins/harness-memory/skills/stateless/reference/official-guidance.md
- 195-198: "CLAUDE_CODE_SKIP_PROMPT_HISTORY ... stops …

harness-ops (26)

plugins/harness-ops/skills/audit-install-state/SKILL.md
- 77-78: "ide/*.lock (its body carries an authToken)": the …
- 202-205: "no value of cleanupPeriodDays touches it" and "the …
- 240-241: "commands/, todos/, statsig/, logs/ being …
plugins/harness-ops/skills/audit-native-overlap/SKILL.md
- 379-382: "A summarizing fetch returns a small model's answer …
plugins/harness-ops/skills/audit-performance/SKILL.md
- 43-47: "doctor (bundled skill, alias /checkup) ... its …
- 160-166: "the sweep pays roughly this walk daily", "/status …
- 217-218: "hooks on one event run in parallel": Claude Code …
- 247-249: "Claude Code reads plugin enablement at startup"
plugins/harness-ops/skills/audit-performance/reference/known-performance-issues.md
- 243-250: "Plugin enablement is read AT STARTUP."
- 687-690: "Exclusions are hidden from non-elevated …
plugins/harness-ops/skills/audit-skill-visibility/SKILL.md
- 59-62: "Claude Code already reports that, through the …
- 129-142: "resolution follows the documented precedence **`local …
- 149-152: "merged per plugin@marketplace key with the …
- 220-229: "SLASH_COMMAND_TOOL_CHAR_BUDGET overrides the whole …
- 256-258: "A user or project skill is listed under its …
- 403-405: "pluginUsage ... counts hook, agent, MCP, and LSP …
- 437-442: "doctor ... finds unused skills, MCP servers, and …
- 484-485: "The native counter is debounced (one write per skill …
plugins/harness-ops/skills/audit-skill-visibility/reference/listing-scorer.md
- 70-78: "the live caller passes the model's window, which is …
plugins/harness-ops/skills/behavior-probes/SKILL.md
- 90-92: "The user CLAUDE.md, auto memory and the account's …
plugins/harness-ops/skills/lanes/context/refresh.md
- 80-85: "${CLAUDE_PLUGIN_DATA} is exported as a real …
plugins/harness-ops/skills/observability/context/data-sources.md
- 119-124: "mcp__ccusage__daily / { since, until } …
- 304-325: "Permission and policy outcomes are emitted as OTEL …
plugins/harness-ops/skills/observability/context/operator-setup-emission-privacy.md
- 38-39: "CLAUDE_CODE_ENHANCED_TELEMETRY_BETA=1 + …
plugins/harness-ops/skills/plugins/context/gotchas.md
- 65-66: "Both show the highest installed version across every …
- 302-303: "The native-Windows jq binary opens stdout in **text …

improvement (4)

plugins/improvement/skills/find/context/ci-health.md
- 66-69: per_page caps at 100 (values above are silently …
- 73-76: conclusion ... run_attempt ... `updated_at − …
- 98-105: the actions_list tooling exposes no created date …
- 19-22: The actions_* toolset (actions_list …

instruction-placement (12)

plugins/instruction-placement/skills/migrate/SKILL.md
- 107-107: "Claude Code matches names exactly; NTFS does not".
- 264-267: "the hook does not fire for an AGENTS.md Claude …
- 475-476: "/init writes CLAUDE.md."
- 482-483: "On a Windows checkout it materializes as a plain text …
- 199-203: "MD022 (blanks around headings), MD024 (duplicate …
plugins/instruction-placement/skills/migrate/reference/shim-droppable.md
- 200-202: "under managed-only no AGENTS.md loads at all."
- 203-209: "a Claude Code version before v2.1.277; the built-in …
- 232-246: "claude-md: no AGENTS.md loads." and …
- 59-63: "the memory page says "your <user-config-dir>/CLAUDE.md" …
- 32-33: "claude-md-or-agents-md (the default) ... Project …
- 100-103: "_ip_reaches bounds the walk at four hops, the …
plugins/instruction-placement/skills/realign/context/apply-recipes.md
- 75-76: "the import inlines at session start and defeats the …

kindle-dedrm (5)

plugins/kindle-dedrm/skills/manage/SKILL.md
- 41-41: Amazon stages 2.9.x installers aggressively; one open …
- 42-70: the cached installer at …
- 75-89: Amazon staged 2.9.x ... …
- 140-142: Calibre KFX Input plugin / Manual via Calibre …
- 154-154: KFXKeyExtractor + KFXArchiver are Windows binaries …

knowledge (13)

plugins/knowledge/README.md
- 100-101: "a sensitive option persists as plaintext on Windows …
- 139-146: "it prints already installed and still writes the …
- 147-150: "Hooks are handed their CLAUDE_PLUGIN_OPTION_* when …
- 167-170: "Plugin option values are read from user …
- 65-74: "PDF works natively with Claude Code's Read tool" …
plugins/knowledge/skills/book-distill/SKILL.md
- 144-144: "git mv invalidates Claude Code's read tracker, so …
- 199-199: "Claude partially reads nested references": owner …
- 65-65: "the fallback covers stock macOS, which ships shasum …
plugins/knowledge/skills/docpage-digest/SKILL.md
- 67-72: "the fallback covers stock macOS, where sha256sum is …
plugins/knowledge/skills/docpage-digest/context/anthropic-docs-profile.md
- 36-38: "Fence attributes arrive as theme={null}." and "URLs …
plugins/knowledge/skills/map-corpus/SKILL.md
- 159-160: "no platform spend-ceiling mechanism exists": owner …
plugins/knowledge/skills/setup/SKILL.md
- 73-80: "Against an already-installed plugin it prints …
- 81-83: "the rendered ${user_config.*} is injected at skill …

machine-health (7)

plugins/machine-health/README.md
- 118-152: "Plugin option values are read from user …
plugins/machine-health/skills/audit/reference/windows/check-catalog.md
- 173-173: "Defender reports AMRunningMode as Passive Mode or …
- 346-351: "LastTaskResult is 267011 …
- 359-360: "Non-elevated, Get-MpPreference returns a literal …
- 669-670: "User Path length >= 2047 (legacy System Properties …
- 232-235: "The script parses the HTML for DesignCapacity and …
plugins/machine-health/skills/setup/SKILL.md
- 144-155: "-s defaults to user"; "A rejected value prints a …

markdown-format (6)

plugins/markdown-format/README.md
- 31-44: "any of the ten documented names"; package.json …
- 45-59: exclude sources; "markdownlint-cli2 applies its own …
- 84-88: config discovery "from the file's directory up through …
- 146-158: "the extensionless …
- 207-210: "supports executable .cjs/.mjs configuration and …
- 271-279: ignores/gitignore key "applies those itself" …

mcp-tools (7)

plugins/mcp-tools/README.md
- 26-32: MCP tool annotations are OPTIONAL in the spec; Claude …
- 39-44: A stdio MCP server does not run in the Claude Code …
plugins/mcp-tools/skills/audit-posture/SKILL.md
- 38-41: the user-level .claude.json holds account data and per-project …
plugins/mcp-tools/skills/audit-posture/reference/checklist.md
- 55-65: Package runners resolve the spec on every server …
plugins/mcp-tools/skills/audit/reference/checklist.md
- 136-156: Python @mcp.tool(meta=...), TypeScript registerTool's …
plugins/mcp-tools/skills/audit/reference/server-discovery.md
- 29-38: SDK spellings for server instructions, delivered via …
- 63-100: Per-SDK tool markers and _meta extraction, including …

miro (4)

plugins/miro/README.md
- 7-12: Claude Code on Windows has a bare-npx spawn bug …
- 86-92: Claude Code caches plugins under …
- 143: Node (>=24) strips TypeScript types at load.
- 237-243: Plugin option values are read from user, --settings …

multi-agent (2)

plugins/multi-agent/README.md
- 87-89: For that command it returns no decision, so the …
- 139-141: the hook cannot start, Claude Code shows a …

mutation-testing (1)

plugins/mutation-testing/skills/audit/SKILL.md
- 211-218: StrykerJS with inPlace: true and mutmut <=2.x write …

overengineering (2)

plugins/overengineering/skills/audit/SKILL.md
- 21-26: "the harness runs a skill's whole pre-compute block as …
plugins/overengineering/skills/delta/context/recurring-wiring.md
- 37-38: "/loop is a bundled skill and needs no install."

performance (1)

plugins/performance/skills/go-faster/reference/catalog/agentic-workflow.md
- 22-22: "third-party skill names are redacted unless …

planning (23)

plugins/planning/skills/plan/SKILL.md
- 23-25: the Bash tool returns the command's complete output …
- 27-32: the harness runs a skill's whole pre-compute block as …
- 42-42: Claude Code's built-in plan mode (shift+tab). That …
- 77-77: Teammates are not worktree-isolated (owner: Claude …
- 234-235: When the reviewer stops at its maxTurns limit its …
- 320-322: /plan [description] enters plan mode ... `/plan …
- 341-343: a research agent that returns a step-by-step approach …
plugins/planning/skills/prd/SKILL.md
- 22-24: the Bash tool returns the command's complete output …
- 26-31: the harness runs a skill's whole pre-compute block as …
- 241-243: There is no supported way for a session to observe …
plugins/planning/skills/interview/context/gotchas.md
- 21-21: plan mode's read-only enforcement blocks it (owner …
- 31-31: There is no way to detect non-interactivity (owner …
plugins/planning/skills/interview/SKILL.md
- 22-24: the Bash tool returns the command's complete output …
- 26-31: the harness runs a skill's whole pre-compute block as …
- 243-245: There is no supported way for a session to observe …
- 341-341: There is no supported way for a session to observe …
plugins/planning/skills/draft-goal-condition/SKILL.md
- 58-58: The evaluator judges the condition against **what …
- 72-72: a fresh-context evaluator model that receives only the …
- 115-121: /goal clear removes it, and bare /goal shows the …
- 145-145: Over-limit submission behavior is undocumented. …
- 147-147: A goal does not change permissions. ... the user still …
plugins/planning/skills/design/SKILL.md
- 20-25: the harness runs a skill's whole pre-compute block as …
- 187-187: Mermaid's own C4 support is experimental (owner …

playbooks (2)

plugins/playbooks/reference/prompt-caching.md
- 145-148: The bundled claude-api skill has a cost-optimize …
plugins/playbooks/skills/repo-sweep/reference/next.md
- 120-122: The Skill tool prints a 'Base directory for this …

playgrounds (1)

plugins/playgrounds/skills/use/SKILL.md
- 92-93: The upstream playground skill writes one HTML file and …

playwright (6)

plugins/playwright/README.md
- 3-10: @playwright/cli offers named sessions, ref snapshots …
- 182-188: Plugin option values are read from user, --settings …
plugins/playwright/skills/check/SKILL.md
- 26: A skill's pre-computed shell injection does not run …
plugins/playwright/skills/demo-video/SKILL.md
- 65-70: Chromium screencast frames stay at CSS-pixel size …
- 89-90: gh pr comment supports --attach in some versions, and …
- 100-104: Kokoro adds about 0.35 s of leading silence per clip.

plugin-quality (10)

plugins/plugin-quality/skills/audit/SKILL.md
- 215-228: An agent's frontmatter model is the dispatch default …
plugins/plugin-quality/skills/audit/reference/component-types/command.md
- 3-5: A command file and a skill both produce /x; skills are …
plugins/plugin-quality/skills/audit/reference/component-types/config.md
- 16-18: Claude Code merges user-global, project and local …
- 27-28: userConfig options marked sensitive: true are stored …
- 31-33: A plugin-root settings.json supports only agent and …
- 34-36: Invalid .lsp.json entries are skipped silently except …
plugins/plugin-quality/skills/audit/reference/component-types/hook.md
- 7: A hook matcher is a regex over tool names.
plugins/plugin-quality/skills/audit/reference/component-types/skill-component.md
- 13-15: A skill's description is the only auto-discovery input …
plugins/plugin-quality/skills/audit/reference/recurring-concerns.md
- 10-11: Hook matchers match on tool name.
- 66-67: .claude/ is not write-protected and not special for …

powershell-format (2)

plugins/powershell-format/README.md
- 21-30: "PSScriptAnalyzer does not auto-discover its settings"
- 46-51: "PSScriptAnalyzer loads and runs those modules' …

rate-limit-guard (7)

plugins/rate-limit-guard/README.md
- 98-111: "A --bg session reports its first prompt as typed" …
- 123-125: "behind a Claude gateway, the spend_limit window"
- 141-143: "Mods can start host processes in the CLI only"
- 219-220: "active once the plugin loads (the next session, or …
- 296-301: "prints already installed and still writes the …
- 313-375: generated How-to-set block (same text as …
plugins/rate-limit-guard/reference/reader-contract.md
- 17-31: .oauthAccount.emailAddress "is **internal CLI …

repo-fleet-hygiene (5)

plugins/repo-fleet-hygiene/skills/audit/SKILL.md
- 143-150: "GitHub's documented 500,000-node ceiling and …
- 124-129: "git worktree list --porcelain (which lists the main …
- 164-166: "global and system Git config, where the transport …
- 336-340: "Windows directory junctions test as symlinks under …
- 377-377: "core.bare=true coincides with working-tree content …

repo-hygiene (11)

plugins/repo-hygiene/README.md
- 162-196: "Plugin option values are read from user …
plugins/repo-hygiene/skills/clean/SKILL.md
- 84-88: "the harness runs a skill's whole pre-compute block as …
- 216-216: "bare, pwsh reads @{...} as splatting syntax and git …
- 79-81: "the Bash tool returns the command's complete output …
plugins/repo-hygiene/skills/clean/context/git-branch-cleanup.md
- 11-11: "ConnectTimeout=5 bounds only the ssh connection. An …
- 49-49: "gh has no unlimited sentinel and rejects `--limit …
- 76-100: "git cherry does not list merges"; "`git branch …
- 196-196: "`git update-ref -d refs/heads/ <captured …
plugins/repo-hygiene/skills/clean/reference/cleanup-config.md
- 28-28: "It runs a full MSBuild evaluation (minutes on a large …
- 58-58: "git clean -fdx traverses directory reparse points …
plugins/repo-hygiene/skills/clean/reference/invocation-forms.md
- 35-38: "reach the model through the Read tool as plain bytes …

review (3)

plugins/review/skills/quality-gate/context/pr.md
- 11-14: the official code-review plugin targets only a PR and …
- 43-46: every code-review plugin run posts a PR comment
plugins/review/skills/security-review/reference/bundled-security-review.md
- 30-31: native /security-review availability depends on its …

ruff-format (4)

plugins/ruff-format/README.md
- 21-31: "a pyproject.toml without [tool.ruff] is ignored …
- 32-34: "ruff check --fix applies safe fixes"
- 41-50: "This extends your config's own unfixable list" …
- 56-57: "--no-cache, so Ruff never writes a .ruff_cache …

session-flow (14)

plugins/session-flow/reference/structure.md
- 234-244: Claude Code compaction replaces the model-visible …
- 262-276: /clear destroys the in-memory task list, and TodoWrite …
- 287-311: TaskCreate takes subject and description, new tasks …
- 700-703: Claude Code stores the session transcript at …
plugins/session-flow/skills/clean-stop/SKILL.md
- 96-99: git status omits ignored files by default, and git …
plugins/session-flow/skills/continue-in-background/SKILL.md
- 108-111: git rev-parse --is-inside-work-tree prints false with …
- 118-121: By default git status shows an untracked directory as …
- 150-160: claude --bg launches a background agent and returns …
plugins/session-flow/skills/reanchor/SKILL.md
- 37-43: FETCH_HEAD holds the tip a git fetch just retrieved …
plugins/session-flow/skills/reconcile/SKILL.md
- 77-83: In Claude Code, a background shell that is still …
- 84-85: A Claude Code subagent's task list is not visible to …
plugins/session-flow/skills/running-retro/SKILL.md
- 43-46: The Claude Code session transcript on disk survives …
- 228-231: Claude Code's /loop runs only while its session is …
plugins/session-flow/skills/show-options/SKILL.md
- 66-68: Claude Code's plugin cache layout is undocumented and …

skill-quality (5)

plugins/skill-quality/README.md
- 110-114: the bundled skill-creator's table-of-contents …
- 248-250: Claude Code prompts for userConfig at enable time and …
- 269-280: reinstall writes config, uninstall drops …
- 297-303: pluginConfigs is read only from user, --settings and …
plugins/skill-quality/skills/setup/SKILL.md
- 72-75: user_config is rendered at skill load

source-control (18)

plugins/source-control/reference/worktree-root-convention.md
- 34-36: ghq list treats worktrees under its root as …
- 46-56: git config --includes defaults off when a scope file …
- 132-148: libgit2 lacks hasconfig:, and JGit and go-git lack …
- 159-164: a wrong SSH signing key verifies Good locally, and …
plugins/source-control/skills/babysit-loop/reference/telemetry-upsert.md
- 8-9: installed plugins cannot reach sibling plugins' scripts
- 13-15: unset user_config keys are left as literal placeholders
plugins/source-control/skills/babysit-prs/reference/orchestration.md
- 46-54: mergeStateStatus reflects required checks, approvals …
- 113-118: mergeStateStatus is recomputed asynchronously and flaps
- 325-330: SendMessage continues an agent; the Agent tool always …
- 573-584: git refuses a branch switch during a merge, with that …
- 799-802: Claude Code's shell cwd can reset between tool calls
plugins/source-control/skills/pull-request/reference/watch-handoff.md
- 10-11: cloud sessions draw on the same plan limits
- 20-21: rate_limits is absent in cloud sessions and under …
- 25-27: /autofix-pr runs in a cloud session
- 31-66: A paused consumer must read the account email from …
- 93-94, 108-111: /autofix-pr resolves the PR from the current branch …
plugins/source-control/skills/worktree/context/audit.md
- 101-103: plugin uninstall -s project has no path flag
plugins/source-control/skills/worktree/fixtures/README.md
- 251-253: jq on Windows emits CRLF, @TSV escapes backslashes …

testing (1)

plugins/testing/skills/write/SKILL.md
- 89: Under Microsoft Testing Platform, --nologo is not an …

toolchain (1)

plugins/toolchain/skills/check/SKILL.md
- 171-180: "go mod tidy -diff needs Go 1.23+"; per-module …

typos-format (3)

plugins/typos-format/README.md
- 37-54: "typos --write-changes applies every correction it …
- 58-61: --force-exclude with `[files] …
- 133-139: "a standalone Rust binary ... (cargo, Homebrew, Conda …

visualization (7)

plugins/visualization/skills/present/SKILL.md
- 105-108: Creating a Slides artifact publishes its title …
plugins/visualization/skills/visualize/SKILL.md
- 116-121: Claude Code substitutes ${user_config.*} values into …
- 186-191: Mermaid renders natively only on a published Artifact.
- 259-262: The Claude Code terminal renders GFM and shows mermaid …
- 287-289: The bundled artifact renderer safely renders 13 stable …
- 342-345: The ClaudeDesign tool lists, gets, writes, copies and …
plugins/visualization/skills/visualize/context/decision-matrix.md
- 225-236: The curated first-party marketplace ships no …

work-items (25)

plugins/work-items/reference/permission-preflight.md
- 97-104: In Claude Code a bare-exact Bash(git commit) …
- 145-151,189-208: Git writes core.worktree only for submodules …
plugins/work-items/reference/tracker-seam.md
- 35-37: jq is not bundled with Git Bash on native Windows and …
plugins/work-items/skills/attend-queue/SKILL.md
- 231-267: A paused lane must read the account email from the …
- 122-125: gh removes the caller as assignee with …
plugins/work-items/skills/setup/SKILL.md
- 99,105-108: git check-ignore without --no-index reports nothing …
- 377-379: Jira Cloud lists resolutions via paginated GET …
plugins/work-items/skills/track/actions/start.md
- 73-107: Git resolves origin/ by trying refs/ …
- 121: In GitHub Actions, gh's @me resolves to the token …
plugins/work-items/skills/work-loop/reference/telemetry-upsert.md
- 58-62: gh api -f body=@file sends the literal string; -F …
plugins/work-items/tools/work-item-tracker/adapters/github/README.md
- 59-64,110-115,141-151,181-184: Cloud sessions refuse REST search/issues as well as …
- 81-84,284-285: gh issue list/label list default to 30 rows silently …
- 95-108: GitHub issue-search qualifiers label:, -label …
- 153-158,174-179: REST issue comments is an integer; gh api lacks --repo …
- 195-201,403-408: Issue comments endpoint defaults to 30/page …
- 211-219: GitHub issue close accepts reasons completed/not …
- 241-244: gh issue edit uses --add/--remove-label/assignee …
- 253-269: REST assignees needs a literal login, and POST returns …
- 317-319: GitHub's issue-closing keywords are …
- 326-334: closedByPullRequestsReferences returns GitHub's …
- 392-402,458: includeClosedPrs:false drops only CLOSED, not MERGED …
- 411-428: GitHub ignores closing keywords in fences/HTML …
- 508-523: gh --json subIssues/parent/blockedBy nodes carry only …
- 524-544: Python open() defaults to locale encoding until 3.15 …
- 545-548: Issue Forms auto-labels only on web-form creation, not …

(repository root) (26)

docs/conventions/prerequisites/README.md
- 19-26: "Claude Code resolves and enforces it" / "Claude Code …
- 151-151: "every POSIX shell sets PPID" (POSIX), "dash as …
- 154-160: "A compaction keeps the session id" -- Claude Code …
- 19-26: Claude Code enforces plugin.json dependencies, keeps …
- 102-104: Claude Code parses a hook's JSON output only on exit 0
- 148: Claude Code runs hook commands in PowerShell on …
- 151-157: compaction preserves the session id
lib/session-bridge/README.md
- 177-186: "CLAUDE_CODE_USE_BEDROCK, _VERTEX, _FOUNDRY …
docs/conventions/hook-budget/README.md
- 60-62: Claude Code mod hooks run in-process as functions and …
- 163-169: A hook that fails to launch or exits 1 is a …
docs/conventions/hook-input-rewriting/README.md
- 51-53: A PreToolUse denial reason reaches Claude identically …
- 83-85: Exit 2 and JSON deny route identically; Claude sees …
docs/conventions/hook-telemetry/README.md
- 42-45: settings.json env values are injected literally with …
- 72: Hook payloads carry prompt_id from Claude Code …
- 86-90: Claude Code's documented hook payload puts session_id …
docs/conventions/plugin-reconfiguration/README.md
- 26-28: claude plugin install --config on an installed plugin …
- 38-55: --config writes pluginConfigs to user settings …
- 57-67: uninstalling a plugin deletes its whole stored …
- 74-78: user_config is rendered at skill load and hook option …
- 79-81: claude plugin install --config with a rejected value …
docs/conventions/routing-as-data/README.md
- 83-84: plugin list ids use the marketplace entry name; the …
docs/extensibility-contract-smoke-tests.md
- 40-49: Claude Code does not validate or absolutize …
- 79-84: CLAUDE_PLUGIN_DATA is per-plugin only in …
- 161-179: multiple:true values render comma-joined, unset keys …
docs/upstream/claude-code.md
- 133-135: CLAUDE_CODE_SUBAGENT_MODEL alone does not change the …
docs/upstream/opus-5-5-task-cost.md
- 223-225: the sonnet alias resolves to an older model on some …

Affected area / component

plugins/attribution (check-stamps.sh), scripts/sync-plugin-options-docs.py, and the plugin and repo docs listed above.

Related: #5684 (no-copy retrofit of files quoting upstream pages), #3465 (provenance sweep).

Found during the upstream-pointer sweep (PRs #7061-#7274).

Thanks for the report. Automated triage handed this issue to a maintainer.

_{Automated triage. A maintainer reviews it.}

This was generated by AI during backlog shaping.

Per the owner's decision on #5684 (comment), the scope #5684 has left folds into this issue. #5684 closes as superseded by the upstream-pointer sweep. Two decisions carry over:

Vendored digest: the skill-authoring playbook body is exempt as a vendored third-party digest. The vendored-skill drift check tracks it (plugins/playbooks/skills/skill-authoring/scripts/update.sh; docs/conventions/upstream-drift/README.md:430-436).
Generators: the plugin-options README block and the hook-event registry are fixed as separate items, one per generator, not inside this issue.

The owner approved this close in an attended backlog-shaping session on 2026-10-17.

This was generated by AI during backlog shaping.

This issue is split into 3 pieces, each sized for one lane PR. There are 525 restated-fact findings, so this cannot be one PR. Two parts are clear and lane-shaped. First, the check-stamps trigger regex misses the inline "recheck when" and "Recheck:" forms (check-stamps.sh:491 on origin/main matches only recheck...trigger, trigger for|is|:, and what would reopen). Second, the copy findings in fable-5-1.md and steps.md are still present on main. Per the backlog-shaping comment, the generator classes go to separate items (#7291 for the plugin-options README block, plus the hook-event registry). The per-plugin restated facts wait on the two open rubric qu

filed: fix(attribution): recognize inline "recheck when" and "Recheck:" triggers in check-stamps (fix(attribution): recognize inline "recheck when" and "Recheck:" triggers in check-stamps #8004)
filed: docs: replace three paraphrased upstream passages with pointer records (docs: replace three paraphrased upstream passages with pointer records #8005)
not filed: chore(attribution): decide the R4 cross-file-record and file-level-stamp readings, then re-slice restated facts per plugin: owner-decision: does a dated record kept in a sibling file, or a file-level default stamp plus a bare pointer, count as a whole record under R4?

Filed pieces are sub-issues of this one, and the paced queue admits them to the lane. Unfiled pieces wait on the blocker shown. The owner approved this split in an attended backlog-shaping session on 2026-10-18.

Decision recorded.

Question: The attribution audit's R4 clears a restated upstream fact when a whole record (pointer, as-of date, recheck trigger) covers it. Judges split on whether two shapes count as that record. Several dozen verdicts depend on the answer. Options: for each reading, yes or no. Reading 1: a record kept in a sibling file. Reading 2: a file-level default stamp plus a bare pointer.

Owner's answer: no to both. R4 is graded "across the containing file", which rules out reading 1. The repo's own rule says "A record never cures a restatement", so a restated value is the defect whatever record sits nearby. Counting either shape would clear findings that rule calls defects. Leave R4 as it is: both "no" answers match its current text, so no rubric version change is needed. Basis: plugins/attribution/skills/audit/reference/rubric.md:337-339, :368-371; .claude/rules/skill-bodies-state-current-rules.md:21-23; docs/conventions/upstream-drift/README.md:79-89.

The owner accepted this recommendation as item D15 of a 17-decision batch in an attended backlog-shaping session on 2026-10-18. The batch question: "Accept all 17 batch 5 decisions as recommended, D1, D2, D3, D5 and D17 included? I post a quoted 'Decision recorded.' comment on each of the 11 issues." The owner answered: "Accept all 17 (Recommended)".

Posted by AI from the owner's answer in that session.

Re-triage.

Sub-issues:

fix(attribution): recognize inline "recheck when" and "Recheck:" triggers in check-stamps #8004 closed (completed) by fix(attribution): recognize inline "recheck when" and "Recheck:" triggers in check-stamps #8165 (7ef5fa7): check-stamps recognizes inline "recheck when" and "Recheck:" triggers.
docs: replace three paraphrased upstream passages with pointer records #8005 closed (completed) by docs: replace three paraphrased upstream passages with pointer records #8138 (605332d): the three paraphrased upstream passages were replaced with pointer records.
The generated plugin-options README block went to Regenerate the plugin-options README blocks with a pointer-only template after the open PRs touching them merge #7291, which is closed (completed).

The third piece was held on the R4 rubric question. The owner answered it on 2026-10-18 (Decision recorded. on this issue): no to both readings. A record in a sibling file does not count, and neither does a file-level stamp plus a bare pointer. "A record never cures a restatement." That removes the hold, and nothing in what is left needs the owner.

The restated facts are still on main. Spot checks on origin/main: plugins/bugs/README.md:96 ("prompted by Claude Code at enable time") and docs/conventions/prerequisites/README.md:156 ("every POSIX shell sets PPID") and :163 ("A compaction keeps the session id"). There are about 525 findings, so intake should split this into per-plugin slices (owner Q93) before any run takes it.

Done when:

/attribution:audit runs report-only on origin/main over every file listed in this issue's "Findings by plugin" section, with R4 as decided on 2026-10-18. A comment here gives the count still present per plugin.
Every finding still present is either rewritten as a pointer to the live upstream section (per .claude/rules/skill-bodies-state-current-rules.md) or removed. This happens in one sub-issue per plugin, or per group of small plugins, each listing its file:line rows.
Each slice's PR passes check-stamps.sh --trigger-less on the files it touches, and each touched plugin gets a changelog fragment.
Rows inside the skill-authoring playbook body are left alone; it is exempt as a vendored digest (2026-10-18 comment, docs: tree-wide no-copy retrofit of files quoting upstream pages #5684 decision).

Report-only audit for #12072: count by plugin.

/attribution:audit ran with no fix on origin/main at 3bd42cabfe6bdef37a05778d396869af0381bb51, with the scripts and rubric from that commit (attribution 0.15.1) and the skill body from the installed 0.15.0, over the 199 files in this issue's "Findings by plugin". R4 applied the 2026-10-18 decision: a record in a sibling file does not count, and neither does a file-level stamp plus a bare pointer. No file was edited. The skill-authoring playbook body is exempt, and no row points into it (0 rows exempted).

Of 525 listed rows: 394 still present (of those, 13 in the generated plugin-options block), 35 gone, 96 not confirmed.

Present: three judges unanimously upheld it under the restated-fact rubric, and it survived the refutation pass.
Gone: the fact is no longer stated in that file (9), it now names a source instead (16), or the panel unanimously cleared it under a carve-out (10).
Not confirmed: the text is still there, but the panel split (12) or the refuter broke the finding (84). Under the skill's rule these stay off the relay and go to no slice.

Plugin	Listed	Present	Of those, generated	Gone	Not confirmed	Sub-issue
actionlint	5	3	0	1	1	#12697
adhd	3	1	0	1	1	#12693
architecture	4	3	0	0	1	#12697
autonomy	14	8	1	2	4	#12688, #12700
bash-format	3	2	0	1	0	#12697
biome-format	1	1	0	0	0	#12698
bugs	13	10	2	0	3	#12690, #12700
code-metrics	3	1	0	0	2	#12698
code-tidying	13	12	0	0	1	#12692
computer-use	7	6	0	0	1	#12694
context-budget	14	9	2	3	2	#12689, #12700
context-guard	10	9	2	1	0	#12694, #12700
context7	3	3	0	0	0	#12697
desktop-notification	2	1	0	1	0	#12698
discipline	5	4	0	0	1	#12686
discovery	24	23	0	0	1	#12687
disk-hygiene	12	7	0	1	4	#12694
docs-hygiene	10	7	0	0	3	#12695
dometrain	12	9	0	3	0	#12693
eol-normalizer	1	1	0	0	0	#12698
evals	18	16	2	0	2	#12691, #12700
event-storming	11	9	0	1	1	#12693
fleet	6	6	0	0	0	#12696
go-format	3	2	0	0	1	#12697
harness-config	55	49	0	1	5	#12685, #12688
harness-memory	9	6	0	1	2	#12696
harness-ops	26	24	0	0	2	#12686
improvement	4	3	0	0	1	#12697
instruction-placement	12	9	0	0	3	#12693
kindle-dedrm	5	2	0	0	3	#12697
knowledge	13	8	0	2	3	#12694
machine-health	7	3	1	2	2	#12697, #12700
markdown-format	6	5	0	0	1	#12687
mcp-tools	7	7	0	0	0	#12695
miro	4	2	0	1	1	#12697
multi-agent	2	2	0	0	0	#12698
mutation-testing	1	1	0	0	0	#12698
overengineering	2	1	0	1	0	#12699
performance	1	1	0	0	0	#12699
planning	23	12	0	4	7	#12692
playbooks	2	2	0	0	0	#12698
playgrounds	1	1	0	0	0	#12699
playwright	6	2	0	1	3	#12698
plugin-quality	10	7	0	1	2	#12695
powershell-format	2	2	0	0	0	#12698
rate-limit-guard	7	6	1	0	1	#12696, #12700
repo-fleet-hygiene	5	4	0	0	1	#12692
repo-hygiene	11	7	1	0	4	#12696, #12700
review	3	3	0	0	0	#12697
ruff-format	4	2	0	0	2	#12698
session-flow	14	7	0	0	7	#12695
skill-quality	5	1	1	1	3	#12700
source-control	18	14	0	0	4	#12691
testing	1	1	0	0	0	#12699
toolchain	1	0	0	0	1
typos-format	3	1	0	2	0	#12699
visualization	7	5	0	0	2	#12696
work-items	25	21	0	0	4	#12689
repository docs (`docs/`, `lib/`)	26	20	0	3	3	#12690
Total	525	394	13	35	96	generated rows: #12700

The per-file lines and the pointer or removal each needs are in the linked sub-issues. The full report is this comment plus those sub-issues.

What this run did not do

Seeded, not nominated. The candidates are this issue's 525 rows, because the question is what is left of them. There was no fresh nomination pass, so a restatement not on the list is not counted.
Panels batched per group of files (DEGRADED). Each judge lens graded about 25 rows in one dispatch rather than one dispatch per row, and one refuter handled each batch. Group A of the original run was marked the same way.
Sources. 200 candidate pages, the corpus ceiling; 164 read through the raw channel, 36 unread. Fingerprinting found no span of 15 words or more on any row's lines against its candidate pages, at the defaults (containment 0.3, span 15), so every row went to the restated-fact panel. A page was matched to a row by URL only, with no page-identity check beyond its first heading, so no tier split is stated. None is fix-eligible.
Outside the row list, not judged. The fingerprint fired on 16 file and page pairs (15 to 28 words, containment at most 0.021) on lines no row names, listed below. Sentences that left a listed file can remain in unlisted ones: the skill pre-compute sentence (r349) is in 32 files.
Moved files. Rows under docs/conventions/<name>/ now read in docs/contracts/<name>/; book-distill is now distill, and show-options is now recommend-skills. context-guard/skills/check is deleted.

check-stamps.sh on the same files: 690 stamps, 2 trigger-less (plugins/context-budget/skills/audit/reference/report.md:37, plugins/planning/skills/design/SKILL.md:393), none expired (window 180 days), none declined.

Not confirmed rows

plugins/actionlint/README.md:48 (r003): refuted
plugins/adhd/README.md:65 (r008): refuted
plugins/architecture/skills/map-flow/SKILL.md:228 (r011): split
plugins/autonomy/skills/setup/SKILL.md:30 (r020): split
plugins/autonomy/skills/setup/context/guardrail-slice.md:56 (r022): refuted
plugins/autonomy/skills/setup/context/trigger-dispatch-slice.md:19 (r023): refuted
plugins/autonomy/skills/setup/context/trigger-dispatch-slice.md:23 (r024): refuted
plugins/bugs/README.md:172 (r034): refuted
plugins/bugs/README.md:166 (r035): refuted
plugins/bugs/skills/write/SKILL.md:117 (r041): refuted
plugins/code-metrics/skills/principles/SKILL.md:60 (r044): refuted
plugins/code-metrics/skills/principles/SKILL.md:173 (r046): refuted
plugins/code-tidying/skills/batch-simplify/context/repo-mode.md:219 (r055): refuted
plugins/computer-use/skills/setup/SKILL.md:123 (r066): refuted
plugins/context-budget/skills/audit/SKILL.md:38 (r074): refuted
plugins/context-budget/skills/audit/SKILL.md:92 (r076): refuted
plugins/discipline/skills/setup/SKILL.md:108 (r098): refuted
plugins/discovery/skills/research/SKILL.md:126 (r116): refuted
plugins/disk-hygiene/README.md:215 (r125): refuted
plugins/disk-hygiene/skills/clean/SKILL.md:73 (r129): split
plugins/disk-hygiene/skills/clean/reference/unsupported-platform-handoff.md:205 (r130): refuted
plugins/disk-hygiene/skills/setup/SKILL.md:170 (r135): refuted
plugins/docs-hygiene/skills/audit-progressive-disclosure/SKILL.md:32 (r138): refuted
plugins/docs-hygiene/skills/extract-ssot/context/anti-patterns.md:129 (r143): refuted
plugins/docs-hygiene/skills/write-for-agents/reference/agent-doc-surfaces.md:139 (r146): refuted
plugins/evals/README.md:142 (r164): refuted
plugins/evals/README.md:143 (r166): refuted
plugins/event-storming/skills/simulation/SKILL.md:61 (r186): refuted
plugins/go-format/README.md:38 (r196): refuted
plugins/harness-config/reference/agents-md-liveness.md:36 (r198): split
plugins/harness-config/skills/audit-automation-gaps/context/hook-timing.md:43 (r199): refuted
plugins/harness-config/skills/audit-instructions/context/phase-a-inventory.md:197 (r206): refuted
plugins/harness-config/skills/audit-permission-state/SKILL.md:211 (r222): refuted
plugins/harness-config/skills/draft-auto-mode-rules/SKILL.md:158 (r249): refuted
plugins/harness-memory/skills/stateless/SKILL.md:65 (r258): refuted
plugins/harness-memory/skills/stateless/reference/official-guidance.md:195 (r261): refuted
plugins/harness-ops/skills/audit-native-overlap/SKILL.md:380 (r265): refuted
plugins/harness-ops/skills/audit-skill-visibility/SKILL.md:256 (r276): refuted
plugins/improvement/skills/find/context/ci-health.md:19 (r291): split
plugins/instruction-placement/skills/migrate/SKILL.md:489 (r295): refuted
plugins/instruction-placement/skills/migrate/SKILL.md:207 (r296): refuted
plugins/instruction-placement/skills/migrate/reference/shim-droppable.md:60 (r300): split
plugins/kindle-dedrm/skills/manage/SKILL.md:75 (r306): refuted
plugins/kindle-dedrm/skills/manage/SKILL.md:140 (r307): refuted
plugins/kindle-dedrm/skills/manage/SKILL.md:154 (r308): refuted
docs/contracts/plugin-reconfiguration/README.md:42 (r310): refuted
plugins/knowledge/skills/distill/SKILL.md:262 (r315): refuted
plugins/knowledge/skills/setup/SKILL.md:83 (r320): refuted
plugins/machine-health/skills/audit/reference/windows/check-catalog.md:414 (r324): refuted
plugins/machine-health/skills/audit/reference/windows/check-catalog.md:303 (r327): refuted
plugins/markdown-format/README.md:125 (r332): refuted
plugins/miro/README.md:10 (r342): refuted
plugins/planning/skills/plan/SKILL.md:22 (r352): refuted
plugins/planning/skills/prd/SKILL.md:21 (r359): refuted
plugins/planning/skills/interview/context/gotchas.md:21 (r362): refuted
plugins/planning/skills/interview/context/gotchas.md:31 (r363): refuted
plugins/planning/skills/interview/SKILL.md:22 (r364): refuted
plugins/planning/skills/interview/SKILL.md:290 (r366): refuted
plugins/planning/skills/interview/SKILL.md:290 (r367): refuted
plugins/playwright/README.md:3 (r378): refuted
plugins/playwright/skills/check/SKILL.md:26 (r380): refuted
plugins/playwright/skills/demo-video/SKILL.md:119 (r383): refuted
plugins/plugin-quality/skills/audit/reference/component-types/config.md:16 (r386): refuted
plugins/plugin-quality/skills/audit/reference/component-types/skill-component.md:13 (r391): refuted
plugins/rate-limit-guard/README.md:143 (r397): split
plugins/repo-fleet-hygiene/skills/audit/SKILL.md:163 (r405): split
plugins/repo-hygiene/skills/clean/SKILL.md:215 (r410): refuted
plugins/repo-hygiene/skills/clean/context/git-branch-cleanup.md:11 (r412): refuted
plugins/repo-hygiene/skills/clean/context/git-branch-cleanup.md:196 (r415): refuted
plugins/repo-hygiene/skills/clean/reference/cleanup-config.md:28 (r416): refuted
plugins/ruff-format/README.md:35 (r423): refuted
plugins/ruff-format/README.md:47 (r425): refuted
plugins/session-flow/reference/structure.md:312 (r426): refuted
plugins/session-flow/reference/structure.md:391 (r428): refuted
plugins/session-flow/skills/continue-in-background/SKILL.md:136 (r431): refuted
plugins/session-flow/skills/continue-in-background/SKILL.md:162 (r432): refuted
plugins/session-flow/skills/continue-in-background/SKILL.md:244 (r433): split
plugins/session-flow/skills/reanchor/SKILL.md:35 (r434): refuted
plugins/session-flow/skills/reconcile/SKILL.md:77 (r436): split
plugins/skill-quality/README.md:127 (r440): refuted
plugins/skill-quality/README.md:307 (r441): refuted
plugins/skill-quality/skills/setup/SKILL.md:78 (r444): refuted
plugins/source-control/reference/worktree-root-convention.md:45 (r446): split
plugins/source-control/skills/babysit-loop/reference/telemetry-upsert.md:9 (r449): refuted
plugins/source-control/skills/babysit-loop/reference/telemetry-upsert.md:31 (r450): split
plugins/source-control/skills/pull-request/reference/watch-handoff.md:60 (r459): refuted
plugins/toolchain/skills/check/context/go.md:51 (r464): refuted
plugins/visualization/skills/present/SKILL.md:108 (r468): refuted
plugins/visualization/skills/visualize/context/decision-matrix.md:227 (r474): refuted
plugins/work-items/skills/setup/SKILL.md:151 (r480): refuted
plugins/work-items/skills/track/actions/start.md:88 (r482): refuted
plugins/work-items/skills/track/actions/start.md:121 (r483): refuted
plugins/work-items/tools/work-item-tracker/adapters/github/README.md:646 (r498): refuted
docs/extensibility-contract-smoke-tests.md:40 (r521): refuted
docs/extensibility-contract-smoke-tests.md:79 (r522): refuted
docs/extensibility-contract-smoke-tests.md:161 (r523): refuted

Fingerprint pairs outside the rows

docs/upstream/claude-code.md against the docs page env-vars
docs/upstream/opus-5-5-task-cost.md against the docs page model-config
plugins/context-budget/README.md against the docs page permissions
plugins/context-guard/README.md against the docs page plugins/mods/reference
plugins/context-guard/reference/reader-contract.md against the docs page model-config
plugins/docs-hygiene/skills/audit-progressive-disclosure/SKILL.md against the docs page skills
plugins/docs-hygiene/skills/audit-progressive-disclosure/SKILL.md against the docs page agent-skills/best-practices (platform)
plugins/dometrain/skills/setup/SKILL.md against the docs page plugins/cli-reference
plugins/harness-config/reference/agents-md-liveness.md against the docs page plugins/mods/overview
plugins/harness-config/skills/audit-pass/reference/doctor-handoff.md against the docs page commands
plugins/harness-config/skills/draft-auto-mode-rules/SKILL.md against the docs page cli-reference
plugins/mcp-tools/skills/audit-posture/SKILL.md against the docs page mcp
plugins/rate-limit-guard/README.md against the docs page plugins/mods/reference
plugins/repo-hygiene/skills/clean/SKILL.md against the docs page skills
plugins/session-flow/skills/continue-in-background/SKILL.md against the docs page env-vars
plugins/work-items/reference/permission-preflight.md against the docs page permissions

This was generated by AI.

Navigation Menu

Repository navigation

Attribution audit: restated upstream facts left after the upstream-pointer sweep, and a check-stamps trigger-form gap #7292

Description

Counts by tier

Top recurring classes

Stamp-detector gap in check-stamps

Open rubric questions

Findings by plugin

Affected area / component

12 remaining items

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Attribution audit: restated upstream facts left after the upstream-pointer sweep, and a check-stamps trigger-form gap #7292

Description

Description

Counts by tier

Top recurring classes

Stamp-detector gap in check-stamps

Open rubric questions

Findings by plugin

Affected area / component

Activity

github-actions commented on Oct 12, 2026

kyle-sexton commented on Oct 18, 2026

kyle-sexton commented on Oct 18, 2026

kyle-sexton commented on Oct 18, 2026

kyle-sexton commented on Oct 28, 2026

12 remaining items

kyle-sexton commented on Nov 3, 2026

Metadata

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Issue actions