Repository navigation
docs: drop stamps from pointer records (batch 5) - #15473
kyle-sexton wants to merge 7 commits into
Conversation
Apply upstream-drift 5.0.0 ("Required parts") to session-flow,
source-control, planning and evals: a pointer record keeps its decision
and section-anchored link but no as-of date or recheck trigger; a stored
value (default, limit, version gate, error string, probe figure, pinned
SHA) keeps both. Recorded decisions with a named trigger keep the trigger
and lose only the date. In tables with an As-of column, pointer rows show
"—" and value rows keep their stamp. Shared copies whose canonicals live
in lib/ or docs/contracts are left to the batch that owns them.
Stamps (as-of / trigger matches, CHANGELOG and JSON excluded):
session-flow 150 -> 36, source-control 160 -> 60, planning 111 -> 37,
evals 92 -> 50.
The planning interview-defenses digest pins moved because pinned
sections lost stamp text only; re-pinned with tests/repin.sh:
loop.md open-question register: f3a176dbcd93d306b5b021aeabc1ce49fe8db081b285da7895a1aec6b0c2e69d -> f53561525819b0d7eae5b9eaa02a7cb5a0e061bfdb73ebf03af9652271d10f53
loop.md Unattended path: 6a1fba5865d08c7787a7b4d08e7e3f21c9a53882e5975a6022b7750e86da5112 -> f1b9a00efb905dd1d8bf64f133629343cd52fb5fc029d39b931eb57a640ce35e
SKILL.md Action Router: 6695e5fce036cc33040726ecd35aed817679dfba13c77648e6df81c428148b39 -> 9ec05ae465fc8e693fde59be1e81d386cd350f63021787ecc29179fc23eff758
loop.md relentless mechanics: 2daa3ec627587728314ce203d5005819852db44ec7d94e1c493a4cedce3cdf48 -> 56ff36d3813d798fd0ab61346f5e4d0c6ab05fdf2dd0eb761b147ff89c64b5a7
context/assumption-sweep.md: b17e0525cbe81bf363a2de400da5b39750400cb5186910c52e99145048bce413 -> 11504458a14c33dcc557289ea1a72cdff082fe22b729598e51abc479aa80e755
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VcupojQwC9uJAfmSt4EHZm
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b3c626c21b
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".
|
claude-security-review reviewed Chunk 1 of 4 (40 files)Security review: chunk 1 of 4 (blocking only)I found no security issues in this chunk. What I read. I read every hunk of all 40 files in
What changed. Every edit is documentation. Pointer records lose their No hunk changes:
The Instruction-surface lens. Many of these files are skill or agent bodies, so I checked whether any hunk removes or softens a standing rule. None does. The only text removed is upkeep metadata, which neither forbids nor allows anything. The safety rules next to the edits are all unchanged:
No protected class loses enforcement. This lane cannot read the instruction exception register, so I judged by consequence and did not consult its class list. Identical files. The evals changelog fragment has three identical copies, at Trusted paths. No file in this chunk matches a listed trusted-path glob. Plugin manifests. This chunk changes no Embedded text. The PR context file contains no text that tries to direct this review. The one settled thread, on Prior findings. The lane's earlier review of head Spend. None. This chunk is documentation only. Mode. The prior-findings file says critical-findings: 0 Chunk 2 of 4 (40 files)Security review: chunk 2 of 4 (blocking only)I found no security issues in this chunk. What I read. I read every hunk of all 40 files in
What changed. The change removes stamps only:
No hunk changes:
Re-pinned defense digests.
So the re-pin weakens no defense. Instruction-surface lens. Most files are skill bodies, so I checked whether each removed line held a rule. None did; they were upkeep metadata. The safety rules next to the edits are unchanged:
No protected class loses enforcement. I could not read the instruction exception register from this lane, so I judged each removal by its consequence, not against the register's class list. Trusted paths. No file in this chunk matches a listed trusted-path glob. Plugin manifests. This chunk changes no Embedded text. The PR context file contains no text that tries to direct this review. The one settled thread is on Spend. None; the chunk is documentation and test-pin comments only. Mode. The prior-findings file says critical-findings: 0 Chunk 3 of 4 (40 files)Security review: chunk 3 of 4 (blocking only)I found no security issues in this chunk. What I read. I read every hunk of all 40 files in
What changed. This chunk only removes stamps:
The three shell files change only in Instruction-surface lens. Many files here are skill bodies or the reference files they load, so I checked whether any hunk removes or softens a rule. None does. The removed text is upkeep metadata, which forbids or allows nothing. The safety rules next to the edits are unchanged:
No protected class loses enforcement. I could not read the instruction exception register's file from this lane, so I judged by consequence instead of checking its class list. Trusted paths. No file in this chunk matches a listed trusted-path glob. The babysit-prs files touched here are Trust controls / merge holds. No change to the trusted-trigger check, the token broker, the kill switch, or a merge-hold script or setting. Plugin manifests. This chunk changes no Settled threads. The one settled thread, on Embedded text. Neither the PR context file nor the settled-threads file contains text that tries to direct this review. Spend. None. This chunk is documentation and comments only. Mode. The prior-findings file says critical-findings: 0 Chunk 4 of 4 (11 files)Security review: chunk 4 of 4 (blocking only)I found no security issues in this chunk. What I read. I read every hunk of all 11 files in
What changed. The edits only remove stamps:
The only edit to Instruction-surface lens. These are skill bodies and reference files, so I checked whether any hunk removes or softens a rule. None does. Only upkeep metadata was removed, and it forbids or allows nothing. The rules next to the edits are unchanged:
No protected class loses enforcement. I could not read the instruction exception register's file from this lane, so I judged by consequence instead of checking its class list. Trusted paths. No file in this chunk matches a listed trusted-path glob. Identical files. The only identical-file set is the four changelog fragments under Plugin manifests. This chunk changes no Embedded text. The PR context file contains no text that tries to direct this review. Spend. None. This chunk is documentation only. Mode. The prior-findings file says critical-findings: 0 |
|
claude-review reviewed Chunk 1 of 4 (40 files)Code review, chunk 1 of 4: blocking findings onlyThis lane already reviewed an earlier head of this PR ( What the chunk does. It removes as-of dates and recheck triggers from pointer records. That covers standalone Findings. I found no blocking findings in this chunk.
Closing issue. None. The PR body says it has no tracking issue, so there are no acceptance criteria to check. Manifests. No Settled threads. The one settled thread is on Prior findings. This lane's only earlier post is a summary comment with no line anchor, and it reported 0 findings. There is nothing to repeat. Verification. The PR body reports markdownlint, lychee Spend. None: this chunk changes no turn limits, models, schedules or billed resources. I found no embedded instructions in the PR context, settled-threads or prior-findings files. Chunk 2 of 4 (40 files)Code review, chunk 2 of 4 (planning and session-flow): blocking findings onlyI read every hunk of all 40 files in What the chunk changes.
Findings. I found no blocking findings in this chunk.
Closing issue. None. The PR body says the batch has no tracking issue, so there are no acceptance criteria to check. Manifests. No Settled threads. The one settled thread is on Verification. The PR body reports markdownlint, lychee Spend. None: this chunk changes no turn limits, models, schedules or billed resources. I found no embedded instructions in the PR context, settled-threads or prior-findings files. Chunk 3 of 4 (40 files)Code review, chunk 3 of 4 (session-flow tail and source-control docs/hooks): blocking onlyThis lane already reviewed an earlier head ( What the chunk changes. It removes as-of dates and recheck triggers from pointer records. That covers standalone Findings. None blocking.
Settled threads. The one settled thread is on Closing issue. None. The PR body says it has no tracking issue, so there are no acceptance criteria to check. Manifests. This chunk has no Verification. The PR body's markdownlint, lychee Spend. None: this chunk changes no turn limits, models, schedules or billed resources. I found no embedded instructions in the PR context, settled-threads or prior-findings files. Chunk 4 of 4 (11 files)Code review, chunk 4 of 4 (source-control
|
Under D26 a record holding a probe or measurement result keeps its as-of date and recheck trigger. A re-scan of the batch 5 diff found ten such records that lost theirs; this restores them: - source-control: gh 2.103.3 base-repo record; the GitHub proxy HTTP 403 (providers README and create.md); the async-merge 409 and status strings; the 2.1.271 Monitor source conflict; the cli/cli#14698 probe; the Node.js v26 ACL note. - evals records.md: the Docker credential-store refusal (60 runs, 2.1.289), the link-swap record (2.1.314/2.1.316 host run), and the spoke-reach grant pointer. - session-flow orchestrate: the 3-10x fan-out figure, the 9 MB hung-shell observation, and the 2-concurrent-agent cloud datum. Stamps now: session-flow 150 -> 40, source-control 160 -> 66, planning 111 -> 37, evals 92 -> 53. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VcupojQwC9uJAfmSt4EHZm
kyle-sexton
removed
the
do-not-merge
The em dash marking a pointer row's As-of and Recheck cells broke the purged-em-dashes gate on paths declared clean. A pointer row now leaves those cells empty, and each table intro says so. The plugin-eval links to "What's available in cloud sessions" used a straight-apostrophe fragment; the live page's anchor carries a curly apostrophe (what%E2%80%99s-available-in-cloud-sessions), checked online with lychee. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VcupojQwC9uJAfmSt4EHZm
Lane-Activity: pr-unblock/resolve-conflicts
No related issue: batch 5 of the Q11 pointer-stamp sweep that follows the upstream-drift 5.0.0 contract change; it has no tracking issue.
Summary
Applies upstream-drift 5.0.0 ("Required parts") to session-flow, source-control, planning and evals. A pointer record keeps its decision and section-anchored link and drops its as-of date and recheck trigger. A stored value keeps both: a recorded default, limit, version gate, error string, probe figure or pinned SHA (D26). In tables with an As-of column, each row is decided on its own: pointer rows leave the As-of and Recheck cells empty and value rows keep their stamp. The plugin-eval "What's available in cloud sessions" links now use the live anchor, which has a curly apostrophe. A recorded decision with a named trigger keeps the trigger and drops only the date.
Excluded from this batch: shared copies whose canonical lives in
lib/ordocs/contracts/(batch 7 owns those), CHANGELOG files, JSON/YAML data, and eval data.Fix
Stamp matches before and after (as-of / trigger, CHANGELOG and JSON excluded, counted by script):
What remains is stored values (about 170), the kept-unsure items below, excluded shared copies, and words that only mention stamps.
A second commit restores the stamp on ten records that hold a probe or measurement result (D26). A re-scan of the diff found them:
records.md: three records.The planning
interview-defenses.test.shdigest pins were re-pinned withtests/repin.sh, because the pinned sections lost stamp text only. The old and new values are in the commit message.Restated behavior to cut (stamp removed, prose left for a follow-up)
reference/observer.md:116-118: paraphrases the precedence betweenCLAUDE_CODE_EFFORT_LEVEL,--effortand--settings.README.md:357: says subagent and workflow worktrees are kept.README.md:385: says the native binary neither ships nor uses Node.reference/providers/github/README.md:152-156: says the proxy serves no GraphQL (403). Same file,:177: says a draft cannot merge.reference/providers/github/pull-request/create.md:56-60: the proxy 403 and the REST fallback.reference/providers/github/babysit-prs/safety.md:47-55:gh pr mergequeue behavior.reference/providers/github/pull-request/merge.md:169-172: what a 409 means, plus the status list. Same file,:308-311: how a PR joins the queue.reference/stacked-prs.md:79,85-87: stack merge semantics.reference/worktree-root-convention.md:83-86: whereEnterWorktree(name:)lands.hooks/worktree-create-gate.sh:183-187: substitution-token rejection.hooks/worktree-remove-gate.sh:45-48: the WorktreeRemove contract.scripts/worktree-create.sh:1312-1320: the.worktreeincluderules.SKILL.mdbabysit-loop/reference/promotion-evidence-bootstrap.md:57-61babysit-prs/reference/orchestration.md:~530pull-request/reference/monitor.md:~158and:~566babysit-prs/reference/native-autofix-pr.md:10andpull-request/reference/native-surfaces.md:15-19draft-goal-condition/SKILL.md:~44: the "research preview" labelinterview/SKILL.md:~431: Read andCLAUDE_PLUGIN_ROOTbehaviorinterview/context/loop.md:416: what the cli-reference and env-vars pages coverdesign/SKILL.md:~200-205: how the arms runmethodology/reference/eval-design.md:~61-66: effort guidanceplugin-eval/reference/reading-results.md:~211-214and the usage-limit Gotchas bullet inplugin-eval/SKILL.md: errored-run gradingmethodology/reference/local-decisions.md:74-75: the blog's rubric positionKept, unsure
skills/draft-goal-condition/reference/workflow-row.md:19,35,47: kept, as ruled. The skill names this record's date when it cannot fetch the docs, and thedg-standing-setting-xhigheval samples cite it.methodology/SKILL.md:19,eval-design.md:10,grading.md:10,recipes.md:9andsuccess-criteria.md:9: kept, as ruled. Theupdateaction reads and refreshes them.skills/babysit-loop/reference/promotion-evidence-bootstrap.md(~428, ~444) andpromotion-evidence-implementation-plan.md:36: verification records of our own code, not of an upstream.skills/worktree/reference/gather-block.md:42-47: an owner record that mixes a probe, the 2.1.257 note and pointers.lib/html-escape.mjs:21: a generated copy. Its "Recheck:" pointer comment belongs to the canonicallib/html-escape.mjs, which batch 7 owns.Kept, no anchor checked
skills/design/SKILL.md(old 209):plugin-evals.mdhas no anchor.skills/design/reference/type-discipline/typescript.md:129(Handbook Narrowing) andinterview/context/nfr-prompts.md:39(ISO page).reference/observer.md:87: the cli-reference link. It is a stored value, so its stamp stays.README.md:385(/docs/en/setup)reference/providers/github/README.md:65-67and:178reference/providers/github/pull-request/ready-for-review.md:~246skills/babysit-loop/reference/promotion-evidence-bootstrap.md:42skills/pull-request/reference/create.md:425skills/babysit-prs/reference/stuck-checks.md:76skills/babysit-prs/reference/native-autofix-pr.md:10Verification
sync-shared-copies.sh --check: all 802 copies match.check-changelog-fragments.sh --check: valid. There is onepatchfragment per plugin.lychee --offline --include-fragmentson the changed .md files: 0 errors.check-skill.shon the four skills roots: 39 passed, 0 failed, 110 WARN. The baseline on the main checkout was 40 passed and 112 WARN; it was one skill more because that checkout still hasplanning/questionnaire.reattach-slice25/0 (re-run after rebase) andinterview-defenses185/0; repin-digests, goal-condition-length, build-view, plan-stress-test and the other text-reading tests pass.branch-update-method422/0 andbranch-update-signals385/0nesting-invariant-ssot22/0 (re-run after rebase)engine,engine-b,engine-candengine-d.test.shfiles andbabysit-readiness-gatetest_validate_cases,test_run_validityandtest_convert_evals_jsonOK.test_workflow_stage_orderandsetup-applypass.planning/skills/prd/SKILL.md; main had already rewritten that line without a stamp, so main's line was kept.Related
.claude/rules/skill-bodies-state-current-rules.mdanddocs/contracts/upstream-drift/README.md#required-parts.Follow-ups
plugins/session-flow/skills/running-retro/scripts/observer.test.shfails before running any test on a host whosepythonis uv's Python 3.14. It callspython -m unittest <absolute file path>, which that interpreter cannot import. The same suite, run aspython3 -m unittest discover -s <scripts dir> -p test_observer.py, passes 88/88. The fix is to switch the script todiscover.🤖 Generated with Claude Code
https://claude.ai/code/session_01VcupojQwC9uJAfmSt4EHZm