Skip to content

docs: drop stamps from pointer records (batch 6b) - #15502

Queued
kyle-sexton wants to merge 3 commits into
mainfrom
docs/drop-pointer-stamps-batch6b
Queued

kyle-sexton wants to merge 3 commits into
mainfrom
docs/drop-pointer-stamps-batch6b

Conversation

@kyle-sexton

Copy link
Copy Markdown
Member

Refs: #15468

Summary

Split 6b of #15475, which was too large for the review lanes to read whole
(MAX_DIFF_BYTES 300000 per judgment, .github/workflows/pr-verify.yml). Drops the as-of date
and recheck trigger from pointer records under upstream-drift contract 6.0.0 (#15468): a pointer
keeps its anchored link; a stamp stays only on a recorded number or string copied or measured from
upstream. Scope: testing, knowledge, code-tidying, disk-hygiene, docs-hygiene.

Fix

The same edits #15475 made to these files, applied on current main; files main changed since
that branch's base were applied as 3-way patches so main's newer text stays. Also drops the stray trigger fragment user-scope keys. left in disk-hygiene skills/clean/reference/safety-model.md, the open review thread on #15475.

Verification

  • Diff: 249741 bytes against main.
  • scripts/check-changelog-fragments.sh --check and --check-required origin/main pass.
  • scripts/sync-shared-copies.sh --check: all registered copies match (run on 7a).
  • Pre-push check-affected passed.

Related

🤖 Generated with Claude Code

https://claude.ai/code/session_01VcupojQwC9uJAfmSt4EHZm

Plugins: testing,knowledge,code-tidying,disk-hygiene,docs-hygiene.
Split from #15475 so the review lanes read the diff whole.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VcupojQwC9uJAfmSt4EHZm
@kyle-sexton
kyle-sexton marked this pull request as ready for review November 12, 2026 02:58
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 12, 2026 •

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

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

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-12T03:03:55.226451Z 02f51cd Draft marked ready
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.

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 02f51cd5b1

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

kyle-sexton and others added 2 commits November 11, 2026 22:09
…nters

The decision tables hold pointers, not stored values, so under contract
6.0.0 they carry the link alone. The probe observation below keeps its
recheck line: it stores the registered description string.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VcupojQwC9uJAfmSt4EHZm
@github-actions

github-actions Bot commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

claude-security-review reviewed 4da638e4969377bd3efa2b5e6ebafe56fe53dc10: read 102 of 102 changed file(s); 0 finding(s), 0 as inline comments.

Chunk 1 of 3 (40 files)

Security review: chunk 1 of 3 (40 files)

I found no security issues in this chunk. I read every hunk of all 40 files. This lane already reviewed an earlier head (02f51cd5), so this review reported only blocking findings, and there were none.

What changed. Every edit is to documentation or comments, matching the pull request description:

  • Each one removes an as-of date, a recheck trigger, or a "relay as of …" note from a pointer record, and keeps the anchored link.
  • The new head also changes both bundled-simplify.md decision tables: they drop their As of and Recheck trigger columns, and the intro text no longer promises those fields.
  • The files are under plugins/code-tidying/, plugins/disk-hygiene/ and plugins/docs-hygiene/, plus one new changelog fragment.

None of these files matches a TRUSTED PATHS glob. No plugin.json or marketplace.json changed, and no workflow, action or merge-gate file changed.

Instruction-surface check. I checked every removed line against the instruction-surface lens:

  • disk-hygiene guard docs (README.md, clean/SKILL.md, safety-model.md, unsupported-platform-handoff.md, fan-out-worker-brief.md, setup/SKILL.md): no rule was removed or softened. These all stay as they were:

    • the kill-switch scope rule (read only from scopes a repository cannot forge) and its fail-closed-to-enabled posture
    • the onFailure: block rows and the exec-form fixed-literal args bound
    • the deny-by-default Bash lane, the belt lifetime and the confirmation gate
    • the rule that AskUserQuestion is preferred and a denied call counts as an unanswered question
    • the warning not to rely on the frontmatter belt inside subagents

    Only stamp text went.

  • clean/SKILL.md frontmatter hook block: only a YAML comment line (# As of: 2026-10-11.) was removed. The hook's type, command and args are unchanged.

  • remove-comments/scripts/change-shape.sh: only the header comment changed. No executable line changed.

  • comment-reviewer.md and compare-labeler.md: only the trailing YAML comment on maxTurns changed. The values (80 and 30), tools and model are unchanged.

  • tidy/reference/exclusions.md: these all stay as they were:

    • the .claude/**, hook-anywhere, .mcp.json, .github/** and frontmatter hard exclusions
    • the hard_exclusions default (enforce)
    • the rule that any other value is read as enforce

    Only the stamp clauses went.

I did not read the instruction exception register (docs/contracts/instruction-exception-register/README.md) in this chunk. I judged the removals by their consequence instead: none of them is a rule that forbids or gates an action.

Identical files. .changes/code-tidying/docs-drop-pointer-stamps-batch6-85a8a9cb.md has copies under other plugins' .changes/ directories. They are all changelog fragments, so a different path does not change what the content does.

Settled thread. The thread on batch-simplify/context/bundled-simplify.md raised a point that is not about security. The owner's fix at this head addressed it, so I did not raise it again.

Embedded text. The PR context and thread text contain no directives aimed at the review lane.

Spend. None. Nothing here changes turns, models, schedules or billing.

critical-findings: 0

Chunk 2 of 3 (40 files)

Security review: chunk 2 of 3 (40 files)

I found no security issues in this chunk. I read every hunk of all 40 files. This lane already reviewed an earlier head, so this pass reported blocking findings only (CRITICAL and IMPORTANT). None came up.

What changed. Each edit drops an as-of date, a recheck trigger, or an inline "as of …; recheck when …" clause from a pointer record and keeps the anchored link. That matches the pull request description. The files are:

  • docs-hygiene: extract-ssot, rename-references, setup, write-for-agents, write-for-humans
  • knowledge: README, work-root.md, absorb-upstream, course-digest, distill, docpage-digest, map-corpus, setup, video-digest, vendor/node-launcher/setup-deps.mjs
  • testing: README, the green-* and test-judge agents, hooks/judge-lib.sh, hooks/test-weaken.sh, and the audit, cleanup and diagnose skills

No file in this chunk matches a TRUSTED PATHS glob. No .claude-plugin/plugin.json or marketplace.json changed, and no workflow, action or gate file changed.

What I checked:

  • Executable files. plugins/testing/hooks/judge-lib.sh, plugins/testing/hooks/test-weaken.sh and plugins/knowledge/vendor/node-launcher/setup-deps.mjs change only in comment lines. judge::read_rule, the noclobber marker line in test-weaken.sh and the install logic are unchanged.

  • Agent frontmatter. In green-fixer.md, green-runner.md and green-verifier.md, only the trailing YAML comment on maxTurns changed. The values (30, 20 and 20), tools, skills and model are unchanged. In test-judge.md, only a YAML comment line changed. tools: "Read, Grep, Glob" and model: inherit are unchanged.

  • Instruction-surface lens. No rule was removed or softened. Every untrusted-source rule is unchanged:

    • the companion-page DATA framing in video-digest/context/companion-primary-sources.md
    • the absorb-upstream "Untrusted source" section
    • the --public-only fetch
    • the credential-handling note in course-digest
    • the Edit-failure abort in rename-references
    • the classifier and skeptic briefs

    absorb-upstream no longer records an as-of date for a page source. Its read, pin and visibility steps are unchanged.

    The instruction exception register was not read in this chunk. I judged each removal by its consequence instead: none of the removed lines forbids or gates an action.

  • Identical files. The changelog fragment set in .claude-lane/identical.txt belongs to chunk 1. Each copy sits under .changes/, so its path does not change what its content does.

  • Settled thread. The one settled thread is on a chunk 1 file and is not a security point. I did not raise it again.

Spend. None. Nothing here changes turns, models, schedules or billing.

critical-findings: 0

Chunk 3 of 3 (18 files)

Security review: chunk 3 of 3 (18 files)

I found no security issues in this chunk. I read every hunk of all 18 files. This was a blocking-only pass, because this lane already reviewed an earlier head.

What changed. Every edit removes an as-of date or a recheck trigger from a pointer record in plugins/testing/ (the diagnose, plan, run-e2e, setup, test-value and write skills and their context files) and keeps the anchored link. That matches what the pull request description says it does. One stale recheck clause in run-e2e/context/e2e.md is trimmed. The bundled-run.md table's As-of cells are left empty on its pointer rows, and its probe rows keep their dates. The only code file is plugins/testing/workflows/fix-until-green.js, where only a comment changed (lines 132–135); no executable line changed. None of these files matches a TRUSTED PATHS glob. No plugin manifest, workflow, action, hook or gate file is in this chunk.

Instruction-surface check. I checked each removed line against the lens by what breaks without it. None of the removed lines forbids, gates or confirms an action:

  • write/context/blind-author.md: the optional deny-rule lock and "the check above stays the gate" are unchanged. Only the stamp and recheck clauses went.
  • run-e2e/SKILL.md: the rule that /verify stays the user's to start is unchanged.
  • run-e2e/context/e2e.md: the failure-packet redaction rules are unchanged.
  • setup/SKILL.md: "Offer it; do not paste it anywhere" is unchanged.

The instruction exception register was not read in this chunk. I judged the removals by their consequence instead.

Settled threads and prior findings. The one settled thread is on a chunk 1 file and raised no security point. The only prior post is the earlier summary, which reported 0 findings, so nothing is restated.

Spend. None. Nothing here changes turns, models, schedules or billing.

critical-findings: 0

@github-actions

github-actions Bot commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

claude-review reviewed 4da638e4969377bd3efa2b5e6ebafe56fe53dc10: read 102 of 102 changed file(s); 0 finding(s), 0 as inline comments.

Chunk 1 of 3 (40 files)

Code review, chunk 1 of 3 (40 files: code-tidying, disk-hygiene, the first part of docs-hygiene, and the changelog fragment)

This review covers blocking findings only. The prior-findings file says this lane already reviewed an earlier head (02f51cd5) and marks the review blocking only: yes. Under that rule I dropped every advisory candidate and posted no new inline comments.

I read every hunk of all 40 files in this chunk.

Result: no blocking findings

The whole chunk is docs or comments. It takes the as-of dates and recheck triggers off pointer records and keeps each anchored link. The one script it touches, remove-comments/scripts/change-shape.sh, changes only in its header comment. No executable line changes.

Nothing in the chunk can fail a build or a check:

  • docs/contracts/upstream-drift/README.md lists "a pointer carries no as-of date or recheck trigger" as Detect-then-judge, not built. No lint enforces stamp shape.
  • The triggers this chunk keeps guard a decision, which the contract allows. Examples: the BSD grep flags in lanes.md, the live probe and the ${user_config…} upstream issues in safety-model.md, and the simplify line-list row in tidy-diff/SKILL.md.

What changed since the earlier round

  • Earlier chunk-1 item 1 is fixed. That item said the decision tables still had a Recheck trigger column. At this head, both batch-simplify/context/bundled-simplify.md and tidy/reference/bundled-simplify.md have two-column Decision | Pointer tables. This matches the owner's resolved reply in the settled-threads file (4da638e).
  • The earlier round's other chunk-1 items are advisory and not restated here. They cover leftover punctuation, a stored value that lost its date, and a reference to a record that no longer has one.

Other notes

  • Identical files: the five changelog fragments under .changes/{code-tidying,disk-hygiene,docs-hygiene,knowledge,testing}/ are byte-identical. Each sits at an ordinary fragment path, so the path does not change what it does. bump: patch fits a docs-only change.
  • Closing issue: none. The PR body says Refs: #15468, so there are no acceptance criteria to check.
  • Plugin manifests: none changed in this chunk.
  • Path-scoped rules: none matched in this chunk.
  • Embedded instructions: none found in the PR context, settled-threads or prior-findings files.
  • Author-claimed verification: the PR body's checks (fragment checks, sync-shared-copies --check, pre-push check-affected) are the author's claims. I did not run them, because this lane has no shell.
  • Cost: none. The change edits docs only and spends no paid resources.

Chunk 2 of 3 (40 files)

Code review, chunk 2 of 3 (40 files: docs-hygiene extract-ssot/context/decision-framework.md through write-for-humans, all of knowledge, and testing through skills/diagnose/SKILL.md)

This review covered blocking findings only. The prior-findings file says this lane already reviewed an earlier head (blocking only: yes), so I left out advisory findings.

I read every hunk of all 40 files in this chunk. I found no blocking findings.

Almost every edit removes a pointer record's As of date and Recheck trigger together. The anchored link stays, and so does the sentence around it. This matches what the PR describes under upstream-drift contract 6.0.0 (#15468).

What I checked:

  • Edits inside a sentence. Where a stamp was cut from the middle of a sentence, the punctuation still works. I checked the closing parentheses in testing/skills/cleanup/SKILL.md:176, testing/skills/audit/SKILL.md (the model-alias pointer) and the distill/SKILL.md Read and Write pointers. I also checked the joined sentences in docpage-digest/SKILL.md "Batch path" and knowledge/skills/map-corpus/SKILL.md Phase 4.

  • Text that describes the record shape. It was updated to match the new shape:

    • write-for-humans/SKILL.md "Source records"
    • write-for-humans/reference/sources.md intro
    • sentence-rules.md "Load"
    • absorb-upstream/SKILL.md intro, the step that records a page URL, and "Report"
    • absorb-upstream/templates/record.md, where the page-source comment now names the pointer record
    • anthropic-docs-profile.md "Hedge preservation"

    The pinned-git-upstream Recheck trigger line in the template stays, as the convention requires.

  • Code files. In knowledge/vendor/node-launcher/setup-deps.mjs, testing/hooks/judge-lib.sh, testing/hooks/test-weaken.sh and the four testing/agents/*.md frontmatter files, only comments changed. Each maxTurns: / model: value and its YAML comment syntax is still valid.

  • Stamps kept on purpose. These are on values copied from upstream:

    • the Playwright v1.63.0 stamps and the stamp on the retries default in testing/skills/audit/SKILL.md
    • the pytest-rerunfailures 15.0 or later stamp
    • the dated probe observation (2026-10-01) in anthropic-docs-profile.md

Below the bar. These are advisory, so this blocking-only review does not count them as findings:

  • plugins/docs-hygiene/skills/write-for-humans/reference/sources.md "Instruction length is a target": the Fable 5.2 pointer lost As of but kept its Recheck trigger. The other records in that file dropped both.
  • docpage-digest/context/batch-path.md and anthropic-docs-profile.md (the .md fetch id probe) each dropped a stamp on what reads like an observed upstream behavior: codex plugin 1.0.6, and a .md fetch missing some rendered ids.

Other notes:

  • Closing issue: none. The PR body says Refs: #15468, so there are no acceptance criteria to check.
  • Plugin manifests: no plugin.json or marketplace.json changes in this chunk.
  • .claude/rules: none of the changed files here matches a path-scoped rule.
  • Settled threads: the one settled thread (batch-simplify/context/bundled-simplify.md) belongs to chunk 1, not this chunk.
  • Embedded instructions: none in the PR context, settled-threads or prior-findings files.
  • Author's verification: the steps in the PR body (fragment checks, sync-shared-copies --check, pre-push check-affected) are the author's claims. I did not run them: this lane has no shell.
  • Cost: none. The change only edits docs and comments and spends no paid resources.

Chunk 3 of 3 (18 files)

Code review, chunk 3 of 3 (18 files: the rest of plugins/testing)

Blocking-only review. This lane already reviewed an earlier head (02f51cd5), so this pass reports only blocking findings and manifest flags. I read every hunk of all 18 files in this chunk.

No blocking findings. Every edit removes an as-of date and its recheck trigger from a pointer record, and each anchored link stays in place, as upstream-drift contract 6.0.0 (#15468) requires. Specific checks:

  • Records that keep their stamp. In run-e2e/context/bundled-run.md, the three pointer rows now have an empty "As of" cell, and the intro says why. The probe rows keep their dates: the 2.1.263 and 2.1.284 binary string searches and the records.json evidence line. That matches the contract's rule that a value copied or measured from upstream keeps its stamp.
  • Triggers kept without a date. Some recheck triggers stay behind with no date: e2e.md "Evidence pointers" and test-value/SKILL.md §1. They guard a research-backed decision, and the contract allows a named trigger in that case. The e2e.md trigger was also narrowed to drop the Playwright-page half, which is now a plain pointer.
  • No broken markdown or punctuation. Merged lines in plan/SKILL.md, write/SKILL.md, setup/SKILL.md and write/context/blind-author.md were checked.
  • Script change is a comment. The edit to plugins/testing/workflows/fix-until-green.js touches only a comment. The code is unchanged and still valid.

Other notes

  • Closing issue: none. The PR body says Refs: #15468, so there are no acceptance criteria to check.
  • Plugin manifests: none changed in this chunk.
  • Path-scoped rules: no .claude/rules changes in this chunk.
  • Embedded instructions: none found in the PR context, settled-threads or prior-findings files.
  • Settled thread: the one settled thread (code-tidying/.../bundled-simplify.md) is in chunk 1, not this chunk.
  • Verification: the PR body's checks (changelog-fragment checks, sync-shared-copies --check, pre-push check-affected) are claimed by the author and unverified. This lane has no shell.
  • Cost: none. This is a docs-only change.

@kyle-sexton
kyle-sexton added this pull request to the merge queue Nov 12, 2026
Any commits made after this event will not be merged.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant