Repository navigation
Conversation
UX Review (Opus 5.5) — ⏭️ skippedRevision |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing blocks. Each item traces to #39277 or to the conductor-mode wording it keeps. The reversed #37881 pin is named in the description, and the linked issue is evidence it was wrong for the Bot. What this change shipsInventory (6 items) — 6 justifiedIntent: stop a Bot's thread filling with status lines on quiet patrol wakes, while finished tasks are still reported in full. FIX.
[FIRST-PRINCIPLES-REVIEWED] ab622a6 Description read[DESCRIPTION-READ] ae4ea4051e2c54c4ba359b118efd25f663b3a7f6b20134402a36ee2d0f4a4cf6 That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review —
|
A Bot's work-ledger patrol wake with no news for the person now ends with nothing_to_do instead of a status line, and a finished task is still reported in full before the patrol stops. Conductor mode keeps its one-line reply: the patrol-reply rule moves out of the shared rule set into a conductor-mode section (also carried by the team lead), and the Bot's conductor.patrol carries its own rule. Closes #39277
fdd6a27 to
ab622a6
Compare
|
Rebased: main moved in docs/system-specs/modules/learn-cron-dashboard.md, docs/system-specs/modules/security.md, src/kiro_crew/cron.py since this head went green (7f8f9d8 -> 5a0389a); CI re-running. The previous head's runs had all been cancelled about a minute after they started, so this push also restarts them. |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS No design-level concerns: the patrol-reply rule moves out of the shared set so each conductor gets the rule its readers need, and the parity allowlist records why the two differ. Notes
[DESIGN-REVIEWED] ab622a6 Description read[DESCRIPTION-READ] 9b896700da40013248134e74739f8dee746323ffca4a14077b3dc230e0b65916 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 |
|
This PR and #39308 both fix #39277 and change the same 7 files the same way. Each moves the one-line patrol reply out of the shared ledger conductor rules in Please keep one of #39308 and #39413 and close the other, carrying over anything only this one has. |
Problem / Motivation
Goal: A Bot's thread stays quiet on a patrol wake with no news, and still reports a finished task in full.
A Bot's thread fills with status lines from its work-ledger patrol: "No completed main run since 05:06Z; the post-fix run is pending.", "Fast Gate green on the fixed heads ... Nothing to act on." Since #37881 the prompt asks for them: the one-line patrol reply is a rule a Bot shares with conductor mode, and it forbids
nothing_to_doon a patrol wake.Why it matters
A person reads a Bot's thread once or twice a day. Every quiet wake that writes a line buries the few messages that need them. Conductor mode is different: a person watches it live, so it keeps its one line.
Not a goal
What changed (motivation → approach → change)
How a conductor talks on a patrol wake depends on who reads it, so that rule is no longer shared.
nothing_to_doautonudge_stopagent_parts/spec_prompts.py: the one-line rule moves out of_LEDGER_CONDUCTOR_RULESinto_CONDUCTOR_PATROL_REPLY, a conductor-mode section the team lead's_SHARED_CONDUCTOR_RULES_BLOCKalso carries (re-exported fromsrc/kiro_crew/agent.py). The sharedwake_checknow says "reply as your patrol-reply rule says", so each side reads its own rule.crewmate_prompts.json: the Bot'sconductor.patrolcarries its own rule ("speak only for news"), and how-you-work items 2 and 5 drop the one-line exception.goal-conductorskill step 6 points at "your prompt's patrol-reply rule" without saying which reply it is.docs/system-specs/modules/crew-mode.mddescribes the split.Backwards compatibility
Compatible: prompt text only; no config key, tool or stored data changes.
Removes nothing: conductor mode and the team lead keep the same one-line rule text, and the Bot only loses an instruction to post a status line.
Tests
test_mate_conductor_parity.py: the old "one line on both conductors" test becomestest_a_patrol_wake_gets_one_line_in_conductor_mode_and_none_on_a_bot(conductor mode and the team lead carry the one-line rule, the Bot carries only its quiet rule, a finished task is reported in full).prompt:patrol_replyis anALLOWLISTdifference with its reason.test_context_composition_contract.py: the twomember_how_you_workpins move to the new text.Proof: PROVEN test/test_context_composition_contract.py::test_stable_text_is_byte_identical[member_how_you_work], test/test_context_composition_contract.py::test_stable_text_is_byte_identical[member_how_you_work_unavailable], test/test_mate_conductor_parity.py::test_every_allowlist_entry_is_still_a_difference (head fdd6a27)
Manual verification
Ran on an isolated pod (signed in, real model): the Bot got a worker task that sleeps 8 minutes, its patrol interval set to 30 s so a quiet wake fires before the task ends. Results are under Real-machine verification.
Real-machine verification
Screenshots / video
N/A -- prompt text only; no frontend file changes.
Related Issues
Closes #39277
Sibling sweep
Swept: git grep -n "one line a patrol wake|patrol wake gets|one exception" origin/main -- src docs/system-specs
Sibling: src/kiro_crew/builtin_skills/goal-conductor/SKILL.md -- changed here
Sibling: docs/system-specs/modules/crew-mode.md -- changed here
Pattern harvest
Rule candidate: review-prompt
Pattern: a rule about how a conductor talks to the person added to the shared rule set, which a Bot and conductor mode both carry, though their readers differ
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...); PR Hygiene does not count commits