[kyle-sexton]

No related issue: batch 3 of the Q11 sweep, which applies the pointer-without-stamps policy from #15449 and owner decision D26; the sweep has no tracking issue.

Summary

This PR applies upstream-drift contract 5.0.0 ("Required parts"), as narrowed by owner decision D26, to plugins/playbooks/: the skills, reference/ and the model-adaptation chapters. CHANGELOG.md and vendor/ copies are not touched.

How each kind of record was handled:

• Pointer records: the as-of date and recheck trigger are removed. The decision and the anchored link stay.
• Stored values (D26): only text that records a number or string copied or measured from upstream keeps its stamp. Examples kept:
  • page bytes, MD5 and SHA-256 in the model-adaptation Sources sections
  • benchmark figures such as SWE-Bench Pro 80.3%
  • the Pocock medians and maximums
  • version gates 2.1.257 and 2.1.259
  • the vendored pin SHA and commit 6fd9479
  • the registry reading in calibration.md (first_party, one non-null id beside seven null)
  • the ultracode, /fast and cost-optimize strings
  • the ~/.claude/keybindings.json path
  • the uv flags --no-build and --no-install-local
  • the managed-agents-onboard command string
  • the version-pinned Claim/Basis catalog rows, which hold flag and action strings
• Stamps removed under D26:
  • described model behavior with no figure
  • system-card findings without a figure
  • "no docs page covers X" notes (the date is gone; the sentence and any trigger stay)
  • chapter-wide triggers with no value under them
  • sync and "last read" dates
• Recorded decisions with a named trigger: the date is removed and the trigger stays. Examples: "a later Fable guide covering the topic", "a live isolated session runs the three forms", "a plugin env var exposes the loaded version".
• Model-adaptation AGENTS.md: the page-conflict convention no longer asks for an as-of date.

Fix

Counts were taken with a script matching **As of**, As of:, Recheck trigger and inline as of YYYY-MM-DD:

plugins/playbooks	Before	After first pass	After D26	After version-probe restore
Stamp lines (as-of plus trigger)	403	167	79	89
As-of date lines	263	87	28	35

Seven probe or read records that name the Claude Code version they were measured on keep their stamps, which were restored after the D26 pass:

• precompute-context.md: three probe records (2.1.296, 2.1.296 and 2.1.316)
• rewrite-a-plugin.md: two observations (2.1.289)
• skill-criteria.md: the prompt-audit source (read at 2.1.292)
• repo-sweep/SKILL.md: the isolation record (read through 2.1.296)

Restated behavior to cut

The prose below is kept and its stamp removed. A later change should cut it.

• skills/boris/reference/orchestration.md ~L53, ~L60, ~L68, ~L75: paraphrase the blog's failure modes, the dynamic-versus-static argument, the six patterns and the use cases.
• skills/boris/reference/autonomy.md ~L354: paraphrases how /goal checks its condition.
• skills/boris/reference/loops.md:17, :30: paraphrase the goal docs (no turn cap; a bare /goal shows spend).
• skills/boris/reference/foundations.md ~L193: paraphrases when PostToolBatch fires.
• skills/boris/reference/customization.md ~L18: paraphrases which terminals take Shift+Enter.
• skills/boris/reference/adoption.md:10: paraphrases the thread's guardrails.
• skills/boris/SKILL.md:82: restates the Ultracode toggle on /effort.
• skills/skill-authoring/reference/precompute-context.md ~L60-61: paraphrases the permission-checks section (a refused command aborts the skill load).
• skills/skill-authoring/reference/skill-criteria.md:13-15: paraphrases the prompt audit's cruft criterion.
• reference/model-adaptation/opus-4-8.md:148, sonnet-5.md:158: paraphrase the Fable guidance on over-prescription.
• reference/model-adaptation/sonnet-5.md:51-52: paraphrases the effort page saying the scale is set per model.
• skills/fable-5/context/verification.md (the Managed Agents automation record): restates the blog's re-check and delivery checks. It keeps its stamp for the command string.
• skills/fable-5/context/context-economy.md:22: restates the keep:"all" behavior.
• skills/repo-sweep/scripts/state.sh:39, tick.sh:22: describe jq.exe text-mode stdout on Windows.

Kept, unsure

• skills/skill-authoring/SKILL.md:34: the post's 9 category names and tip titles are strings copied from the post. Whether a whole-skill restatement counts as a stored value is unsettled.
• Sources read dates that have no figure beside them. They are a read log rather than a stamp form:
  • opus-4-8.md ~L150-152
  • haiku-5-5.md:244, :246
  • sonnet-5-5.md ~L286, ~L289
  • sonnet-5.md:183-186
  • opus-5.md:185

Kept, no anchor checked

• skill-criteria.md:387
• prompt-caching.md:43
• haiku-5-5.md:208
• sonnet-5-5.md:264

Out of scope

lib/prerequisites.{sh,mjs} and scripts/fetch-docs.sh are shared copies of the root lib/ canonicals.

docs/contracts/upstream-drift/README.md:152-156 still asks a "no docs page covers" record to keep its date. That text predates D26, which overrides it, so the contract needs its own follow-up.

Verification

• markdownlint-cli2 on the changed markdown: 0 issues.
• check-skill.sh on plugins/playbooks/skills: 29 WARN before and after both passes; 5 passed, 0 failed.
• sync-shared-copies.sh --check: all 802 copies match.
• check-changelog-fragments.sh --check: valid.
• lychee --offline over plugins/playbooks/{reference,skills}: 0 errors.
• No eval grader or test greps for a removed stamp.
• Review thread on fable-5/context/orchestration.md:86 (asking to restore a "no docs page covers" date): answered with D26 and resolved.

Related

• docs: pointers carry no as-of date or recheck trigger #15449 (the policy this applies)

🤖 Generated with Claude Code

https://claude.ai/code/session_01VcupojQwC9uJAfmSt4EHZm

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-11T23:28:39.002995Z	5d1c121	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review reviewed 58840497bf290b7ec255aa5611c4f54f025ad250: read 3 of 3 changed file(s); 0 finding(s), 0 as inline comments.

Security review (incremental): PR #15461, docs: drop stamps from pointer records (playbooks)

I found no security issues in this push. This lane already reviewed 0b084b92b37273c2c79132cc14bd5000ceab1d88, so this review was blocking only: it reports only CRITICAL and IMPORTANT findings.

Scope. I read every hunk of all 3 files listed in .claude-lane/incremental-files.txt, from .claude-lane/incremental.diff:

• plugins/playbooks/skills/repo-sweep/SKILL.md
• plugins/playbooks/skills/skill-authoring/reference/precompute-context.md
• plugins/playbooks/skills/skill-authoring/reference/skill-criteria.md

None of them matches a TRUSTED PATHS: glob.

What changed. The push restores the version-probe stamps that the PR description lists under "After version-probe restore":

• As of lines that name a Claude Code version (2.1.296, 2.1.316, 2.1.292).
• Their Recheck trigger lines.

In repo-sweep/SKILL.md, the recheck trigger gets wider: besides the original "a live isolated session runs the three forms", it also fires when the docs section or a release note changes. Every hunk only adds lines, except two one-line replacements, and each of those keeps the original text and adds to it.

Instruction-surface lens. These files are a skill body and skill references. Nothing in this push deletes, narrows or softens a rule. Every line it adds is a date, a version or a recheck condition, and none of them allows, forbids or gates an action. I didn't consult the protected-class register because no rule was removed.

Other surfaces. This push touches no executable code, manifest (plugin.json or marketplace.json), workflow, token, trust-control, kill-switch or merge-hold file.

Prior findings and settled threads. This lane's only earlier post is the summary comment from the last review, which reported 0 findings, so nothing carries forward. This push doesn't touch the settled thread on fable-5/context/orchestration.md:86.

Embedded text. The PR context contains no instructions aimed at the review lane.

critical-findings: 0

[github-actions]

claude-review: 1 inline finding(s) on 5d1c12180b5cc93713dc40477a450f0c59efcab2; the summary is in this lane's pull request comment.

claude-lane: claude-review

[github-actions]

claude-review reviewed 58840497bf290b7ec255aa5611c4f54f025ad250: read 3 of 3 changed file(s); 0 finding(s), 0 as inline comments.

Code review: PR #15461, incremental review (0b084b9 → 5884049)

Scope: I read every hunk of all 3 files in .claude-lane/incremental.diff. I used the full diff and the PR context for background only. This lane already reviewed earlier heads, so this review covers blocking issues only. Advisory findings were left out.

Result: no blocking findings. This increment puts back the version-probe stamps that the PR description lists under "After version-probe restore":

• repo-sweep/SKILL.md:49-51: the isolation record gets back As of: 2026-10-11 (section and release notes through 2.1.296). Its trigger is widened to cover a change in the docs section or a release note, and it still includes "a live isolated session runs the three forms". This matches the body text at L41-43, which says the forms were read through 2.1.296.

• precompute-context.md: three probe records get their stamps back:

  • L118-120 (2.1.296)
  • L200-201 (2.1.296)
  • L226-229 (2.1.316)

  Each stamp names the Claude Code version the probe ran on, which fits D26's version-gate class of stored value.

• skill-criteria.md:377: the prompt-audit source gets back As of 2026-10-06 and a recheck trigger, beside its read at 2.1.292 value.

I checked the list structure at the head commit around each edit, and the bullets still render correctly.

The description also lists two restored observations in rewrite-a-plugin.md (2.1.289). That file is not in this increment, so this review did not look at it.

Settled thread: the owner answered the thread on fable-5/context/orchestration.md:86 with decision D26, and those lines have not changed since. I did not raise it again.

Closing issue: the PR closes none (it is batch 3 of the Q11 sweep), so there were no issue criteria to check. I found no embedded instructions in the PR context, the settled threads or the prior findings.

Manifests: this increment changes no plugin.json or marketplace.json.

Verification: the markdownlint, check-skill, sync-copies, changelog and lychee results in the PR description are the author's claims. I did not run them.