Repository navigation
docs(rfc): wake a conductor when a waiting worker's wake_at comes due - #28896
Conversation
Adds rfc-work-report-wake-at.md (status accepted, repository owner's decision 2026-11-03) and its index row. A progress work_report may carry an optional wake_at (ISO-8601 UTC, at most 24h ahead). The work-ledger probe raises one wait_due wake when it passes, with a timestamp compare and no network call, and the item is not stale before it. Docs only; the implementation follows in its own pull request.
UX Review (Opus 5.5) — ⏭️ skippedRevision |
First Principles Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: this PR only adds documentation, and the RFC covers the root cause (no time field, plus a stall token that fires only once) with a timestamp check that never calls the network. Notes
[DESIGN-REVIEWED] d764cdc Description read[DESCRIPTION-READ] ef00bf6068c8ebe01c0ced9924e73580eef5a09a103010ffcdbf78a0122f0abf That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
Problem / Motivation
Goal: Record the owner's decision to let a waiting worker say when it should be resumed, so the conductor is woken at that time instead of hours later.
A worker that waits on a time (for example a GitHub rate-limit reset) reports
progress, which never wakes the conductor. The one stall wake per report comes 15 minutes after the report, usually before the wait is over. After that the ledger reads idle and the patrol waits 40 quiet ticks, about 6.7 hours at the 600 s interval. On 2026-11-02 this left a worker idle that long. Workers must not usewait, because it holds a turn on the shared kirocrew-core pool.A research pass over main (re-checked at
41bade9256) found:progressis left out ofWORK_WAKE_STATUSESon purpose (src/kiro_crew/work_vocab.py).is_actionable_eventinsrc/kiro_crew/ledger_wake.pyrefuses it on its status alone.src/kiro_crew/probes/work_ledger.pyspends the tokenstall:<last_report_at>. A spent token never fires again while the item is open.PATROL_MESSAGEinsrc/kiro_crew/conductor_patrol.pysays "Ignore progress", and the goal-conductor skill mapsprogressto "nothing"._IDLE_LEDGER_QUIET_FLOOR(40 ticks) insrc/kiro_crew/autonudge_service/gate.pysets the 6.7-hour gap.Why it matters
A conductor running several workers loses hours of progress every time one waits on a rate limit or a CI run. The work sits finished-but-parked until a person notices or the floor turn comes. The new field changes the worker contract and a crew-log entry, so GOVERNANCE.md asks for an RFC on main before the code.
Not a goal
What changed (motivation → approach → change)
docs/request-for-change/rfc-work-report-wake-at.md, statusaccepted(the repository owner chose this design on 2026-11-03).docs/request-for-change/README.md.The design: an optional
wake_at(ISO-8601 UTC) on aprogresswork_report, refused if in the past or more than 24 h ahead. It is stored on the ledger item and the crew-log entry, and shown inwork_ledger_read(compact too) and on the goal board. Each tick the work-ledger probe compares one timestamp, with no network call and no model turn. When the time comes it raises onewait_duewake, deduped per(last_report_at, wake_at). An item is not flagged stale before itswake_at. The patrol text, goal-conductor skill and worker prompt change to match.Alternatives in section 6: a worker-armed one-shot
monitor_start(the wait stays invisible on the board and fails silently when the worker forgets), and re-waking the conductor on stale (a model turn per re-wake, and it undoes #18112's saving). Section 7 estimates about 80 to 120 non-test lines and lists the files.Backwards compatibility
Compatible: docs only; nothing that works today changes.
Removes nothing: the diff only adds a document and an index row.
Tests
Proof: not provable -- docs or config only
scripts/docs-lint.shpasses locally.Manual verification
N/A — docs only; every file and symbol the RFC cites was re-checked on main
41bade9256.Related Issues
Follow-up implementation PR will reference this one.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributed under the project's Apache-2.0 license.