Repository navigation
Record the landscape sweep's measured usage (stopped run no longer 'unknown'; completed run's effort verified) - #154
Merged
seathatflowsinourveins merged 2 commits intoSep 24, 2026
Conversation
…is known, the completed run's effort verified The stopped first run's usage was recorded as unknown; agent-lab's child-usage.mjs measures it from the retained transcripts (17 children, claude-sonnet-5 at effort high, 168,052 output tokens). The completed run's per-request usage and resolved model/effort (all 119 children at effort max) are added beside the runtime's own counter, labelled as a separate counter. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
seathatflowsinourveins
enabled auto-merge (squash)
September 24, 2026 01:21
Codex Review SummaryThis 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: b79949e052
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".
…the stopped run as a lower bound Codex review of #154: sanitized child-usage.mjs output for both runs is retained with the tool's commit and sha256, the command, its exit code and the raw output's sha256. The stopped run is recorded as incomplete: its totals are lower bounds and the unobserved remainder is unknown. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
seathatflowsinourveins
deleted the
claude/sweep-usage-accounting-20260924
branch
September 24, 2026 01:46
9 of 22 tasks
seathatflowsinourveins
added a commit
that referenced
this pull request
Sep 24, 2026
… plugin-install commands, dated dispositions (#161) * Apply the 2026-09-24 community sweep's catalog adoptions - Effort guard: Claude Code discards SessionEnd hook output, so a self-heal now also appends a one-line notice that the next SessionStart reporting a model shows once and deletes (atomic claim by rename, 0600, O_NOFOLLOW; still exits 0 and never blocks). Every existing precedence rule and message is kept. SHA256SUMS updated; tests cover the two-event handoff, single consumption, the model gate, one combined JSON output and an unwritable notice path. - Community practice page: correct the devcontainer row, which credited sandbox/permission settings that do not exist, and add the dated 2026-09-24 disposition rows for twelve repositories with pointers to the keep-but-compare items. - Recipes: context-mode's 6f0cc68 is a reviewed revision, not an enforced pin. A Claude marketplace source takes a branch or tag, and the former @<sha> form exited 1 on Claude Code 2.1.281 in a scratch config, so the command uses the documented no-ref form. claude-hud's row gains its v0.8.0 tag commit. - Bootstrap step 4a: new-PC check of installed_plugins.json gitCommitSha against the recipe rows (added after v2026.09.23.1), bound to the rows by a drift test. - docs/decisions/2026-09-24-community-sweep.md records all 69 items: adoptions and where each applies (the credential-export change is pending its owner), 14 keep-but-compare rows with their named measurements, 40 rejections with reasons, and the sweep's gaps. - Re-register the changed files' hashes in manifests/evidence.json. test_shipped_guard_is_verbatim fails on a host until its installed guard is refreshed from this commit (tools/adoption/install_claude_profile.py --only guard). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * Resolve the community-adoption review: guard notices, plugin commands, records - Effort guard: SessionStart resolves its own warning before it claims any notice (a settings file it cannot parse costs the warning, never a notice), claims each notice by atomic rename, and deletes the claimed file only after its unbuffered stdout write succeeds; a failed write renames the claim back. Each SessionEnd notice is now its own file, written under a temporary name and renamed into ~/.claude/effort-default-guard.notices/, so no append can land in a claimed file. A start that loses the claim race prints nothing for that notice, and notices, claimed files and temporary files older than 7 days are deleted unseen. The guard still exits 0. SHA256SUMS updated. - Guard tests cover each path: a lost claim race and a notice written during a claim (os.rename patched in a driver process), eight simultaneous starts, a failed print, malformed settings, stale markers and a symlinked notice. Each new test failed against a targeted mutant. test_shipped_guard_is_verbatim compares with the host copy only outside CI and when that copy exists; otherwise it skips with a reason. - Plugin commands: the handbook and the explorer's token topic drop the `claude plugin marketplace add ...@<sha>` form, which exits 1, and the "pinned" wording; the handbook adds the installed_plugins.json gitCommitSha check against the reviewed revision. Bootstrap step 4a gives the recipe's install commands, and its check reads $CLAUDE_CONFIG_DIR. - Docs-consistency test: the plugin-revision mutants go through the same checkers as the real assertions (bootstrap and recipe commit drift, a renamed key, a missing commit, command drift, a missing install), and a new scan fails any tracked document that gives a Claude marketplace add a commit. - A15 evidence: evidence/artifacts/community-sweep-20260924/ plugin-marketplace-refs.json retains the raw native runs, registered in manifests/evidence.json. On Claude Code 2.1.281, same repository: commit ref exit 1, tag ref and no ref exit 0. codex-cli 0.155.1 --ref checks out 6f0cc68. The step 4a check ran against a scratch install and this host. - Records: agent-lab adoptions are marked pending merge of agent-lab branch claude/community-adoptions-20260924 (6a297c7), and every deferred item has a named owner. A14 now cites #154. The star-audit rows are described as unchanged, with an owner for the pointer, and the pin-file note stays, with its reason. The follow-up sweep (obra/superpowers, ruvnet/ruflo, Anthropic primary sources) is recorded. - Rehash the changed registered files in manifests/evidence.json. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * Race the notice-claim test for real; correct the sandbox, Explore and agent-lab pointer wording - test_simultaneous_starts feeds all eight guards from barrier-synchronised threads, so they actually contend for the claim (5/5 runs pass). - The project has no Claude Code sandbox settings but does ship Codex ones. - Explore: 31 of the 34 built-in children, those spawned without a per-call model. - Agent-lab pointer moves to 4831e53. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: seathatflowsinourveins <234074349+seathatflowsinourveins@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A follow-up to #153, correcting a recorded fact.
The stopped first sweep run's usage was recorded as unknown. agent-lab's
.claude/workflows/child-usage.mjsmeasures it from the retained subagent transcripts: 17 discovery children (9 returned, 8 stopped in flight),claude-sonnet-5at effort high, 168,052 output / 302 input / 8,233,147 cache-read / 773,809 cache-creation tokens. This is now recorded in the attempt record and in the lane limit.The completed run's per-request usage is added beside the runtime's own counter and labelled as a separate counter, not to be summed with it. The same data verifies the effort claim: all 119 children resolved to
claude-sonnet-5/claude-opus-5-5at effort max, with 4,225,386 output / 3,544 input / 138,862,781 cache-read / 9,733,523 cache-creation tokens.Changes. Two
lane_limitsstrings inmanifest-20260923.json, the attempt record, and theirmanifests/evidence.jsonhashes. No candidate, receipt or count changes, and nothing that enters a lane packet.Checks.
validate.py,evidence_manifest --check,landscape.py,build_verdicts --checkand the gate all pass, along withtests.test_sota_convergence. Guarded gitleaks: no leaks.🤖 Generated with Claude Code