Repository navigation
docs(playbooks): fix the 53 drift findings from the held-files sweep #12106
Description
Problem
The last wave of the repo-mode /multi-agent:audit-defaults sweep (#12067, recorded on #6916) confirmed 58 findings in the 49 files held back from earlier waves. The sweep's own PR fixes the 5 in plugin-quality and mutation-testing. This issue holds the other 53: 13 in harness-config and 40 in playbooks.
Each finding was upheld by its skeptic panel (the votes column is upheld-refuted) and re-checked against raw upstream pages fetched on 2026-11-01. No quote was contradicted. Not fully checked:
boris/reference/customization.md:63quotes in single quotes, which the check does not search.boris/reference/orchestration.md:33and:147rest onmodels/overview, which is not in the docs index, so it was not read. Line numbers are at origin/maine09dba5d92; re-read each line first.
Fix each finding the way .claude/rules/skill-bodies-state-current-rules.md says: our decision in our own words, plus a pointer to the exact section with an as-of date and a recheck trigger. Never put restated upstream text beside the pointer. The playbooks/reference/model-adaptation/ chapters also follow that directory's AGENTS.md.
These change behavior and go through #11823's eval-gated route, not a pointer PR:
harness-config/.../audit-instructions/reference/criteria.md:467: removesopus-4-8from the inert set.criteria.md:1337: changes the per-message effort carve-out.audit-instructions/SKILL.md:251: re-pins the bytes-per-token budget.
Findings
| file:line | kind | votes | upstream | proposed disposition |
|---|---|---|---|---|
plugins/harness-config/skills/audit-instructions/reference/criteria.md:467 |
stale | 3-0 | build-with-claude/prompt-engineering/prompting-claude-opus-4-8 | Take opus-4-8 out of the inert set and keep opus-5. Add pointers with as-of 2026-11-01: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5#controlling-subagent-spawning, https://platform.claude.com/docs/en/b... |
plugins/harness-config/skills/audit-instructions/reference/criteria.md:1337 |
stale | 3-0 | build-with-claude/effort | Point the exclusion at https://platform.claude.com/docs/en/build-with-claude/effort#per-message-effort-beta and stop naming only between_tools. As currently written, the carve-out also has to exclude Haiku 5.5 with thinking: {"type": "disabled"}. That c... |
plugins/harness-config/skills/audit-instructions/reference/criteria.md:1422 |
stale | 3-0 | model-config | Either delete the v2.1.111 version fence, or re-pin it to a live source that states it, with a fresh as-of date. The pointer record says the env-vars row no longer states the fence, so the record no longer matches its source and the fence cannot be rechecked. |
plugins/harness-config/skills/audit-instructions/reference/criteria.md:1038 |
stale | 3-0 | sub-agents | Re-read the three anchors and drop or restate the recorded 'Source conflict', then re-date the pointer to 2026-11-01. The pointer text is truncated, so this rests on what is visible: the omitClaudeMd row and the --agent passage now agree. |
plugins/harness-config/skills/audit-instructions/reference/criteria.md:1730 |
copied-current | 3-0 | build-with-claude/prompt-engineering/prompting-claude-opus-4-8 | Add a pointer: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-4-8#calibrating-effort-and-thinking-depth (or effort#recommended-effort-levels), as of 2026-11-01. Recheck trigger: the Opus 4.8 guide stops recomm... |
plugins/harness-config/skills/audit-instructions/reference/criteria.md:1466 |
copied-current | 3-0 | build-with-claude/effort | Add a pointer to the model-agnostic source: https://platform.claude.com/docs/en/build-with-claude/effort#effort-with-tool-use, as of 2026-11-01. Recheck trigger: that section stops tying higher effort to more tool calls. The claim's 'agentic search and codi... |
plugins/harness-config/skills/audit-instructions/reference/criteria.md:1437 |
copied-current | 3-0 | build-with-claude/effort | Replace the bare '(effort page)' with https://platform.claude.com/docs/en/build-with-claude/effort#effort-with-thinking, as of 2026-11-01. Recheck trigger: that section starts treating effort as a thinking budget or as a substitute for budget_tokens. |
plugins/harness-config/skills/audit-instructions/reference/criteria.md:1265 |
copied-current | 3-0 | workflows | Complete the pointer record with the URL https://code.claude.com/docs/en/workflows#ask-for-a-workflow-in-your-prompt and an as-of date (2026-11-01). The quoted record names the section and a recheck trigger but gives no address and no as-of date. Consider a... |
plugins/harness-config/skills/audit-instructions/reference/criteria.md:462 |
copied-current | 3-0 | build-with-claude/prompt-engineering/prompting-claude-fable-5-1 | Add pointers for both arms, as of 2026-11-01: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5#parallel-subagents for fable-5 and https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/promp... |
plugins/harness-config/skills/audit-instructions/SKILL.md:251 |
stale | 3-0 | about-claude/glossary | Re-pin the 3.5 bytes-per-token setting. The glossary entry this pointer relies on now limits the ~3.5-characters figure to models before Claude 4.7, and says 4.7 and later produce about 30 percent more tokens for the same text. Every current alias resolves ... |
plugins/harness-config/skills/audit-instructions/SKILL.md:174 |
copied-current | 3-0 | model-config | Either change the example to a placeholder (e.g. "opus currently resolves to <version>") or add a pointer to https://code.claude.com/docs/en/model-config#model-aliases with an as-of date and a recheck trigger (the alias table or the Version history row ... |
plugins/harness-config/skills/audit/reference/audit-checklist.md:257 |
stale | 3-0 | sub-agents | Change the wording and add a pointer to https://code.claude.com/docs/en/sub-agents#choose-an-effort-level and #choose-a-model with an as-of date. A definition's model:/effort: pair is no longer all that a spawn runs on. A per-invocation effort paramet... |
plugins/harness-config/skills/audit-instructions/SKILL.md:348 |
copied-current | 3-0 | sub-agents | Add a pointer to https://code.claude.com/docs/en/sub-agents#how-forks-differ-from-other-subagents with an as-of date and a recheck trigger (that table changes what a fork inherits). The independence argument rests on upstream fork behavior that is stated co... |
plugins/playbooks/reference/model-adaptation/opus-4-8.md:79 |
stale | 3-0 | build-with-claude/effort | Re-point the second link to https://platform.claude.com/docs/en/build-with-claude/effort#recommended-effort-levels (the Claude Opus 4.8 entry) and refresh the as-of date to 2026-11-01. The '#recommended-effort-levels-for-claude-opus-4-8' heading is gone. Th... |
plugins/playbooks/reference/model-adaptation/opus-4-8.md:82 |
stale | 3-0 | build-with-claude/effort | Re-point the effort link to https://platform.claude.com/docs/en/build-with-claude/effort#recommended-effort-levels (Claude Opus 4.8 entry) and refresh the as-of date to 2026-11-01. The per-model anchor no longer exists. The max_tokens guidance still matches... |
plugins/playbooks/reference/model-adaptation/opus-4-8.md:156 |
copied-current | 3-0 | model-config | Add a pointer record. Point to https://code.claude.com/docs/en/model-config#work-with-fable for the Fable side and https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-4-8#more-literal-instruction-following for the ... |
plugins/playbooks/reference/model-adaptation/opus-5-5.md:207 |
stale | 3-0 | fast-mode | Keep the body and the pointer (https://code.claude.com/docs/en/fast-mode#decide-when-to-use-fast-mode still supports it). Re-pin the recheck trigger: on Opus 5.5, fast mode is already out of research preview, so the 'fast mode leaves research preview' trigg... |
plugins/playbooks/reference/model-adaptation/sonnet-5-5.md:258 |
stale | 3-0 | fast-mode | Re-pin the recheck trigger. Fast mode for Opus 5.5 is now generally available, so 'fast mode leaving research preview' has already partly happened and no longer works as a single event. Replace it with 'the supported-model list in the fast mode page's openi... |
plugins/playbooks/reference/model-adaptation/sonnet-5-5.md:15 |
copied-current | 3-0 | model-config | Keep the rule but stop restating the fallback target inline. Replace 'if the session moves to Sonnet 5' with 'if the session falls back to another model (see the fallback section below / https://code.claude.com/docs/en/model-config#automatic-model-fallback)... |
plugins/playbooks/reference/model-adaptation/sonnet-5.md:51 |
copied-current | 3-0 | model-config | Keep the pointer and the as-of date, and fill in the empty recheck trigger. Example: 'a re-read of model-config#choose-an-effort-level or effort#recommended-effort-levels-for-claude-sonnet-5 no longer saying levels are calibrated per model, or a change to S... |
plugins/playbooks/reference/model-adaptation/sonnet-5.md:40 |
copied-current | 3-0 | model-config | Keep the claim and pointer, and add the missing recheck trigger. Example: 'the pointed sections no longer present effort as the control for depth of reasoning on Sonnet 5'. The as-of date (2026-08-04) is older than the effort page's current Sonnet 5.x lineu... |
plugins/playbooks/reference/model-adaptation/sonnet-5.md:172 |
unpointed-judgment | 3-0 | model-config | Add a pointer for the stronger-model half to https://code.claude.com/docs/en/model-config#work-with-fable, with an as-of date and a recheck trigger such as 'that section stops advising outcome-level, larger tasks for Fable'. Back the weaker-model half with ... |
plugins/playbooks/reference/model-adaptation/sonnet-5.md:169 |
unpointed-judgment | 3-0 | model-config | Add a pointer for 'over-prescription degrades Fable' to https://code.claude.com/docs/en/model-config#work-with-fable. Point the 'helps you' (Sonnet 5) half at the Sonnet 5 prompting guide section that supports it. Record an as-of date and a recheck trigger ... |
| file:line | kind | votes | upstream | proposed disposition |
|---|---|---|---|---|
plugins/playbooks/skills/boris/SKILL.md:123 |
stale | 3-0 | workflows | Reword: ultracode is a prompt keyword that opts one prompt into a workflow, and it is also a session setting. Drop "guaranteeing", because there are named cases where the keyword does not start a run. Point at https://code.claude.com/docs/en/workflows#ask... |
plugins/playbooks/skills/boris/SKILL.md:121 |
copied-current | 3-0 | workflows | Keep the row, which is correct, and add a pointer to https://code.claude.com/docs/en/workflows#ask-for-a-workflow-in-your-prompt with an as-of date of 2026-11-01. Recheck trigger: that section changes which phrases opt in. |
plugins/playbooks/skills/boris/SKILL.md:109 |
copied-current | 3-0 | model-config | Keep the pointer and add the missing record: an as-of date (2026-11-01) and a recheck trigger (the Adjust effort level table changes which models accept xhigh or their defaults). The pointer is present, but the row lacks the date and trigger the other effor... |
plugins/playbooks/skills/boris/SKILL.md:66 |
copied-current | 3-0 | model-config | Keep the pointer to https://code.claude.com/docs/en/model-config#choose-an-effort-level and add an as-of date (2026-11-01) and a recheck trigger (a new model release, or that section changes its levels or defaults). |
plugins/playbooks/skills/boris/reference/customization.md:35 |
copied-current | 3-0 | model-config | Replace the copied five-level list with a pointer to https://code.claude.com/docs/en/model-config#adjust-effort-level, or keep it and add that pointer, an as-of date (2026-11-01) and a recheck trigger (a new model release or a change to the levels table). T... |
plugins/playbooks/skills/boris/reference/customization.md:63 |
copied-current | 3-0 | sub-agents | Add a pointer to https://code.claude.com/docs/en/sub-agents#frontmatter-reference with an as-of date (2026-11-01) and a recheck trigger (the frontmatter table changes). The field list is correct but partial: effort, memory, skills, hooks, mcpServers, isolat... |
plugins/playbooks/skills/boris/reference/customization.md:65 |
copied-current | 3-0 | sub-agents | Add a pointer to https://code.claude.com/docs/en/sub-agents#frontmatter-reference, or to the settings-reference agent entry, with an as-of date (2026-11-01) and a recheck trigger (the agent setting or the --agent flag is renamed). |
plugins/playbooks/skills/boris/reference/foundations.md:60 |
copied-current | 3-0 | model-config | Keep the Superseded banner and its pointers, and add an as-of date (2026-11-01) and a recheck trigger (the Extended thinking section changes which models can turn thinking off). The banner lacks the as-of date and trigger that its sibling banner (a1c33) car... |
plugins/playbooks/skills/boris/reference/autonomy.md:186 |
stale | 3-0 | permission-modes | Replace the plan list with a pointer to https://code.claude.com/docs/en/permission-modes#eliminate-prompts-with-auto-mode, and record an as-of date (2026-11-01) and a recheck trigger (the requirements list in that section changes, or the section is renamed ... |
plugins/playbooks/skills/boris/reference/autonomy.md:165 |
copied-current | 3-0 | model-config | Keep the Superseded pointer to https://code.claude.com/docs/en/model-config#adjust-effort-level, which still resolves and still states per-model defaults, but add an as-of date (2026-11-01) and a recheck trigger (that section stops listing per-model default... |
plugins/playbooks/skills/boris/reference/autonomy.md:172 |
copied-current | 3-0 | model-config | Same as a2c1: keep the Superseded pointer to model-config#adjust-effort-level and add an as-of date and recheck trigger. 'xhigh is the new default' is true only for Opus 4.7. The current opus alias (Opus 5.5) defaults to medium, so the Superseded note has t... |
plugins/playbooks/skills/boris/reference/autonomy.md:174 |
unpointed-judgment | 3-0 | model-config | Replace 'leave it alone for most work' (said of xhigh) with a pointer to https://code.claude.com/docs/en/model-config#choose-an-effort-level, plus an as-of date and a recheck trigger. Current guidance says the model's own default suits most work, and that d... |
plugins/playbooks/skills/boris/reference/autonomy.md:237 |
copied-current | 3-0 | model-config | Point at https://code.claude.com/docs/en/model-config#adjust-effort-level for the level list, with an as-of date and a recheck trigger (the levels table changes). Note that available levels depend on the model. The five-level scale is correct for current mo... |
plugins/playbooks/skills/boris/reference/autonomy.md:233 |
copied-current | 3-0 | model-config | Add a pointer to https://code.claude.com/docs/en/model-config#adaptive-reasoning-and-fixed-thinking-budgets with an as-of date and a recheck trigger (the list of always-adaptive models changes). The statement is correct, and the source now says it applies t... |
plugins/playbooks/skills/boris/reference/autonomy.md:253 |
copied-current | 3-0 | build-with-claude/prompt-engineering/prompting-claude-opus-5-5 | Keep both pointers, which still resolve and support the amendment. Add the missing as-of date (2026-11-01) and a recheck trigger (model-config#extended-thinking stops listing which models can't turn thinking off, or the prompting-guide section is renamed or... |
plugins/playbooks/skills/boris/reference/autonomy.md:292 |
unpointed-judgment | 3-0 | none | Either point this Opus 4.7 behavior claim at the Opus 4.7 prompting or migration guidance it came from, with an as-of date and a recheck trigger, or label it as 4.7-specific history. No fetched source supports it as current guidance, and the current opus al... |
plugins/playbooks/skills/boris/reference/orchestration.md:95 |
stale | 3-0 | workflows | Drop 'guarantees'. Point the keyword sentence at https://code.claude.com/docs/en/workflows#ask-for-a-workflow-in-your-prompt (and #where-the-keyword-works), with an as-of date and a recheck trigger of 'that section changes the keyword or where it applies'. ... |
plugins/playbooks/skills/boris/reference/orchestration.md:160 |
stale | 3-0 | model-config | Mark this Fable-era note as historical, or reword it so Fable is not called the default for coding. Point at https://code.claude.com/docs/en/model-config#work-with-fable and https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-int... |
plugins/playbooks/skills/boris/reference/orchestration.md:33 |
stale | 3-0 | model-config | Change the banner to say Fable 5 was the strongest model as of Jun 9, 2026 and has since been succeeded. Add a pointer to https://code.claude.com/docs/en/model-config#work-with-fable (or https://platform.claude.com/docs/en/models/overview#compare-models), a... |
plugins/playbooks/skills/boris/reference/orchestration.md:147 |
stale | 3-0 | about-claude/models/overview | Re-pin the anchor to https://platform.claude.com/docs/en/models/overview#compare-models. The #latest-models-comparison anchor no longer exists. The pricing pointer (#model-pricing) still resolves and can stay as it is. |
plugins/playbooks/skills/boris/reference/loops.md:17 |
stale | 3-0 | goal | Remove 'or the turn cap is reached', since there is no built-in turn cap. Say instead that the goal clears when it is met, judged impossible, or hits an error the user has to fix, and that a turn or time bound has to be written into the condition. Add a poi... |
plugins/playbooks/skills/boris/reference/orchestration.md:123 |
stale | 3-0 | skills | Drop 'Experimental: treat as preview, not stable API', or mark it as the tip's own launch-time caveat. The live docs present context: fork as a regular documented field. The pointer record can stay. |
plugins/playbooks/skills/boris/reference/loops.md:29 |
copied-current | 3-0 | goal | Add a pointer record: https://code.claude.com/docs/en/goal#check-status for the spend shown by a bare /goal, and https://code.claude.com/docs/en/workflows#cost for per-agent token usage in /workflows. Include an as-of date and a recheck trigger of 'either s... |
plugins/playbooks/skills/fable-5/context/calibration.md:72 |
stale | 3-0 | models/fable-5-1/overview | The 2026-08-03 local-registry half of this record is past its own recheck trigger: a Claude model has been released since. Re-run the registry read against the pinned Claude Code build (2.1.316) and check whether Mythos 5 and now Mythos 5.1 are still unsele... |
plugins/playbooks/skills/fable-5/context/calibration.md:68 |
stale | 3-0 | build-with-claude/thinking-troubleshooting | Re-pin the pointer from #supported-models to https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations, the anchor the page itself uses for the per-model table. Then re-record the size and MD5. The table now has ... |
plugins/playbooks/skills/fable-5/SKILL.md:2 |
copied-current | 3-0 | model-config | Keep. A frontmatter description cannot carry a link. Add the description's model list to the scope of the SKILL.md line-19 pointer record (model-config#available-models and #work-with-fable, as of 2026-11-01), so that record's recheck trigger also covers th... |
plugins/playbooks/skills/skill-authoring/reference/skill-criteria.md:203 |
stale | 3-0 | sub-agents | Keep the pin rule, but fix the fallback sentence. An omitted field now falls back first to CLAUDE_CODE_SUBAGENT_MODEL when it is set, and only then to the main conversation's model. Also say that CLAUDE_CODE_SUBAGENT_MODEL_FORCE=1 ignores frontmatter `model... |
plugins/playbooks/skills/skill-authoring/SKILL.md:206 |
stale | 3-0 | skills | Add the fork exception to the model paragraph. With context: fork, an excluded value follows the subagent model-override rules: a blocked family alias moves to the newest version the allowlist permits instead of simply not being used. The sentence in a5... |
plugins/playbooks/skills/skill-authoring/reference/skill-criteria.md:291 |
copied-current | 3-0 | sub-agents | The list is correct today but has no pointer. Either point to https://code.claude.com/docs/en/sub-agents#choose-a-model for the alias set, or keep the list and add as-of 2026-11-01 with the recheck trigger 'the Agent-tool model alias list changes'. |
plugins/playbooks/skills/skill-authoring/reference/skill-criteria.md:60 |
unpointed-judgment | 3-0 | about-claude/models/optimizing-for-cost-and-intelligence | Either cite a basis for 'weaker models lean on these terms more' or reword it as a hypothesis the per-model measurement tests. The closest upstream guidance says something different: instructions added for older models get over-obeyed by newer ones. That su... |
Done when
- Every row above is fixed in a PR or dropped after a re-read, with the reason on this issue.
- The three behavior-changing rows go through docs(skills): ship the five eval-gated #6916 drift fixes after a no-loss measurement #11823's eval-gated route.
- Each new pointer carries an as-of date and a recheck trigger.
Activity
github-actions commentedon Nov 1, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Large
Complexity: Medium
Uncertainty: Knowable
Flags: none
Blocked by: none
Understood: 53 drift findings from the held-files sweep (13 in harness-config, 40 in playbooks). Each needs a pointer with an as-of date and a recheck trigger, or a reword. Three rows change behavior and go through #11823's eval-gated route. The work is mostly docs-only, so it gets low priority. It touches many files across two plugins, so it should be split into per-plugin or per-file-group PRs. Each row must be re-read first, because line numbers are pinned to an older commit. No duplicates found among open issues.
Automated triage, applied by the lanes App.
melodic-automation-lanes commentedon Nov 1, 2026
Not admitted to the issue lane yet: sub-issue #12108 is open.
melodic-automation-lanes commentedon Nov 1, 2026
Split at intake (owner Q93). This issue is Size Large, so its done-when is split into 2 sub-issues, each holding its task-list lines as written.
Automated split by the lanes App. Admission waits on this issue while a sub-issue is open; work-unblock closes it as completed once every sub-issue is closed as completed by a merge or a listed human and its done-when is covered.
melodic-automation-lanes commentedon Nov 4, 2026
Every sub-issue of this split is closed as completed by a listed human or a merge (#12107, #12108), so the lanes App closes it as completed.
github-actions commentedon Nov 4, 2026