Repository navigation
claw: claw-t27f/suggest/refresh pass with a note when the App token cannot be minted; claw-refresh can read the queue (#10576) - #10585
Merged
Merged
Conversation
… App token cannot be minted; claw-refresh lists PRs 25 a page (#10576) claw-refresh was red on every push to master: its GraphQL listing asked for 100 open PRs with mergeStateStatus a page, and with 207 open the second page outran GitHub's timeout (an HTML 502, "invalid character '<'"). 25 a page lists all 207 in the same time. claw-t27f, claw-suggest and claw-refresh now have the #10568 shape: the token step is continue-on-error, a failure step writes the owner's fix to the step summary, and later steps run only with a token. claw-batch is operator- dispatched and stays red on purpose, marked `# token_shape: red` for the law in specs/claw/token_shape.t27 (#10577). Shrink offset: scripts/mcp-wrapper.sh (-10), ported to specs/port/scripts/mcp-wrapper.t27 and sealed; nothing runs it. Refs #10576 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag
enabled auto-merge (squash)
October 18, 2026 17:58
2 of 3 tasks
gHashTag
disabled auto-merge
October 18, 2026 18:09
Closed
3 of 4 tasks
…most (100 - queued) / 25 pull requests (#10588) With the GraphQL listing healed, claw-refresh would update up to 30 behind pull requests after every push to master. Each moved head starts about 25 workflow runs, so one merge could queue 750 runs while 886 were already waiting. The master ruleset is not strict, so a refresh only buys freshness: it never unblocks a merge. specs/claw/refresh.t27: - refresh_budget(queued) refreshes nothing at QUEUE_ROOM (100) or more queued runs. Below that it allows (QUEUE_ROOM - queued) / RUNS_PER_REFRESH (25), never more than REFRESH_CAP. An unread count refreshes none. - refresh_plan_queued is what the job calls. It adds one summary line saying which case applied. - refresh_plan_within carries the old body with the cap passed in; refresh_plan keeps its meaning (REFRESH_CAP). - Six new cases: a budget sweep over -1..1000, a busy queue, a quiet queue, an unread queue, a limit above the cap, and the text pins. - Lab results: 50/50 passing, 0 vacuous. Three negative controls fail their cases as they should: cap not lowered, room check removed, unread branch removed. gate-topology.yml claw-refresh: - reads actions/runs?status=queued total_count (permissions gain actions: read) before the listing loop; - passes the count to the harness, which parses it strictly (empty, non-numeric, negative or > i32 is unread); - these inputs were run through the harness on the lab: 886, 0, 30, empty, abc, -, -3 and 99999999999. gen/c/claw/refresh.c regenerated; seal claw_ClawRefresh.json remade. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag
changed the title
gHashTag
enabled auto-merge (squash)
October 18, 2026 18:15
2 tasks
…ried listing) #10559 landed the same queue gate and listing fix this branch carried (refresh_budget, first:25), with more: three tries, a cap of 5 and only PRs the merger lands next. specs/claw/refresh.t27, gen/c/claw/refresh.c and the seal are master's. This branch keeps the token shape for claw-t27f, claw-suggest and claw-refresh, the claw-batch red marker, and adds actions: read to claw-refresh: without it the queue read falls to -1 and the job refreshes nothing, silently. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag
changed the title
This was referenced Oct 18, 2026
Merged
Merged
4 of 5 tasks
This was referenced Oct 18, 2026
This was referenced Oct 18, 2026
Closed
3 of 5 tasks
This was referenced Oct 18, 2026
Contributor
This was referenced Oct 18, 2026
gHashTag
added a commit
that referenced
this pull request
Oct 18, 2026
…ecs, one run set (#10652) * claw: queue_cost.t27 -- job-time share and queued-run classes, so the queue measurement is reproducible (Closes #10600) The 2026-10-18 queue measurement (#10590) was taken by hand with jq and awk. This spec is that measurement: cost_plan sums job seconds per workflow and names the top holder (t27b native, 88% in the fixture); class_plan sorts queued runs into current head, superseded head, closed PR, draft PR, other event and unreadable. Writing it found a defect in the published table: the 97 draft runs were counted twice (94 on a current head, 3 on a superseded one), so the classes summed to 1019, not 922. One row is one class here; the fixture sums to 922. 13 tests pass under t27c test-report and under t27b; five mutations (floor rounding, draft before superseded, no sha check, no seconds bound, empty name) each fail at least one test. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * claw: push_gate.t27 -- skip a fix push that only restarts required checks (Closes #10606) Every push to a PR head restarts its required checks at the back of the Actions queue. push_gate decides each lane push first: hold a new PR while more than QUEUE_LIMIT runs are queued, imported from refresh.t27 rather than copied; push a fix only when a required check is red or the PR's own diff causes the red; skip it when master already fixed the job, as on #10585. 14 tests, t27c and t27b pass, six mutations each fail, gen-c and gcc clean. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs(claw): competitors addendum -- merge queue, Kodiak and push_gate on when checks run again (Refs #10606) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * claw: seal evidence, evidence_pins and suggest_feed -- every claw spec sealed, 24 of 24 (Refs #10608) All three pass t27c test-report and t27b (12, 4 and 19 tests, none vacuous); their seals were the only ones missing in specs/claw. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(claw): check_state.t27 reads a queued check as pending, not failing (#10617) The PR Dashboard's per-PR line counts a check as failing when its conclusion is not SUCCESS, SKIPPED or null. A queued check has conclusion "" and a commit status has none, so #10585 read "21 failing" with no failure, and a failed commit status was never counted. check_state.t27 classifies each rollup row (status TAB conclusion, or CONTEXT TAB state) as pass, fail, wait or unreadable, and gives one verdict: red, pending, none or green. Unreadable is never green; no checks is none (#3248). The dashboard's current rule is kept as a named function so the #10585 measurement is a test. t27c 12/12 with 0 vacuous, t27b 12/12, five mutations each fail a test, gen-c and gcc clean, seal saved. queue_cost.t27 exports qc_line_end and qc_split for it; its seal is re-saved. Refs #10617 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(claw): check_latest.t27 -- a re-run supersedes the run it replaces (#10617) check_state.t27 read every rollup row. On 2026-10-19 #10563 held "Issue Gate/check-linked-issue" twice: run 39151562346 COMPLETED FAILURE and its re-run 39157315158 QUEUED. check_state called that red; the check that decides the merge is the queued re-run, so it is pending. check_latest.t27 takes rows with the check's name and run id. Per name, the largest run id (compared as a number) is live; older rows are superseded. A commit status is never superseded. A malformed row is unreadable and supersedes nothing. Live rows are classed by check_state. check_state's MAX_ROWS is renamed CHECK_MAX_ROWS: queue_cost also exports MAX_ROWS, and a spec importing both lost the name in both compilers. t27c 11/11 with 0 vacuous, t27b 11/11, six mutations each fail a test, gen-c and gcc clean; check_state still 12/12; seals saved and verified. Refs #10617 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(claw): required_red.t27 -- push_gate's req_red flag read from the rollup (#10606) push_gate.t27 takes req_red from its caller, and the lane set it by eye. On 2026-10-18 that eye pushed #10585 for a red claw-sarif, which is not a required check. required_red.t27 reads check_latest's rows (name, run id, status, conclusion). A row is required when its job is exactly validate, check-linked-issue, parse-ratchet or own-language. Only the newest run per name counts. req_red is true when a live required check fails, unknown when any row is unreadable, false otherwise. A missing required check is counted and is not red. t27c 9/9 with 0 vacuous, t27b 9/9, five mutations each fail a test, gen-c and gcc clean, seal saved; the five branch seals verify. Refs #10606, #10617 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(claw): merge_lane.t27 -- the merger's starvation read from its own runs (#10644) TRI CLAW's merger job waits in the same ubuntu-latest pool as every pull_request job. On 2026-10-18 it had 34 runs queued, the oldest 6598 s old, and one job waited 57 min for a runner. ClawMergeLane reads one row per merger run (status TAB conclusion TAB age) and prints counts, the oldest open age, and F merger idle|flowing|starved|unknown (starved: an open run older than 1800 s; a known starve wins over an unreadable row). t27c 8/8 with 0 vacuous, t27b 8/8, 7 mutations caught, gen-c clean (the duplicate-const warning is gen-c's, #10645), seal saved. Refs #10393 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(claw): merge_lane yield -- the merger's runs merged 0 PRs (#10644) 39 of the last 40 merges on 2026-10-18 had GitHub's native auto-merge enabled (auto_squash_enabled in the timeline); the merger workflow is not on the merge path. Its last 10 successful runs merged nothing: nine workflow_run runs each took a runner to log "no bee-reviewed label". yield_plan reads one row per merger run (event TAB merged TAB no_label TAB other) and prints a Y line and F merger-yield zero|some|unknown|none. t27c 15/15 with 0 vacuous, t27b 15/15, 6 new mutations caught (none BLOCKED), seal re-saved, 6 branch seals verify. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(claw): merge_lane share -- the merger is 11% of the queue (#10644) On 2026-10-19, 23 of the repository's 209 queued runs were the merger's (22 of them workflow_run), the largest single group, while its last 10 runs merged nothing. share_plan reads one row per queued run (workflow TAB event) and prints an S line and F merger-share low|high|unknown|none; high at 10% or more, counting each unreadable row as somebody else's. t27c 23/23 with 0 vacuous, t27b 23/23, 8 new mutations caught (none BLOCKED; the eighth needed a new test), seal re-saved and verified. gen-c compiles with warnings (#10645). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <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.
Summary
App-token shape (#10576). Three jobs now follow the #10568 pattern: claw-t27f, claw-suggest and claw-refresh.
continue-on-error.steps.tok.outcome == 'failure'step writes the owner's fix to the step summary.if: steps.tok.outputs.token.claw-t27f's latest scheduled run was red for exactly this reason: "The level of access for permissions requested are not granted to this installation" (checks write).
claw-batch stays red on purpose. An operator dispatches it with a list of PRs, so a token that cannot be minted must fail that run. The token step carries the
# token_shape: redmarker, which the law in #10577 reads.claw-refresh gets
actions: read. #10559 added the queue read. The job'spermissions:block has noactionsscope, so if the read is refused the count falls to -1 and the job refreshes nothing, silently. The header comment now gives #10559's limits: a cap of 5 and a queue limit of 150, not 30.Superseded part. This branch also carried its own claw-refresh listing fix and queue budget (#10588). #10559 landed the same fix first, with more: three tries, a cap of 5, and a refresh only for PRs the merger lands next. The merge takes master's
specs/claw/refresh.t27,gen/c/claw/refresh.cand seal unchanged.Run over this branch,
specs/claw/token_shape.t27names only claw-sarif, which #10568 fixes. Once both merge, it names nothing.Shrink offset.
scripts/mcp-wrapper.shis deleted (−10). It is ported tospecs/port/scripts/mcp-wrapper.t27and sealed, and nothing runs it. The workflow is +16/−13 against master, so foreign lines move +16/−23. The census does not countrun:inside a flow mapping (#10569).Refs #10576, #10588, #10545
🤖 Generated with Claude Code