Repository navigation
docs(rfc): propose per-goal limits, drift signals and a frozen done check - #37649
Conversation
First Principles Review (Opus 5.5) — ⏭️ skippedRevision |
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS This is a docs-only draft RFC, and its storage claim checks out: Notes
[DESIGN-REVIEWED] 437f171 Description read[DESCRIPTION-READ] 05ff3f92f08976b322de51fffa9656d0446b843c0e0c1a5190d9c6b39c52d314 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: |
Code change approver recommendation: A person should decideWhy: the decider estimates a 10% chance that this change introduces a defect, within the 12% limit for approving on its own, but it changes 404 lines, more than the 400 it may approve on its own. This PR adds documentation only: a new draft RFC ( Trust boundary. The RFC adds owner-only fields to a Bot's goal record. Its claim about where that record lives checks out: the Citations. The ones I spot-checked match the head commit:
CI. All checks are green, and every automated review lane passed or was skipped. Remaining issues. The internal inconsisten An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit |
c256114 to
437f171
Compare
|
This adds a new concept, so it needs a design sign-off from a maintainer before it can merge. A maintainer will post the decision here. Review result: no blocking findings on this head. The PR changes docs only, a new draft RFC and one README index row, so no code or test rule applies. We checked the security claims against the code: Optional:
|
bolichen97
left a comment
There was a problem hiding this comment.
Review input on head 437f1715b0f5. This is an RFC, so this is a comment, not an approval.
The core move is sound: every limit and signal refuses work in the store with a code, so none of them depends on the Bot following a prompt rule. The rollout phases have exit criteria a test can fail. docs-lint.sh and test/test_pr_quality_gates.py (97 passed) are clean locally, and all CI checks pass on this head.
Things to fix in the text
- Stale baseline. Since
audited-at: 2041e8681c, #37381, #37648 and #37651 have merged, and the branchcrew/slack/goal-cost-on-goals-tabis now #37665, merged today. §2.2 rows 1, 10, 11, 12, §5 and the README row still call these open or unshipped. Re-audit against current main and bumpaudited-at. - Self-contradiction (also raised by the Design Review). The header lists
rfc-work-ledger-concurrency-cap.md§7.2 under "Builds on, and does not change", but open decision 7 recommends amending it. Either move §7.2 out of that list or make decision 7 conditional. - §4.2 step 2 refuses a
session_createfrom "the Bot's chat or its workers" when a goal's signal fires.session_createis not tied to a goal, so a Bot with two goals, or an owner asking the Bot for something unrelated, would also be stopped. Say how a session is attributed to a goal, or scope the refusal to the goal's board and its workers only.
Open decisions
- Default lines. I would not stop by default on every signal at once. Phase 1 already measures the counts; ship stop-by-default only for signals whose source phase 1 confirmed and whose line held on real goals, and show the rest only. A busy but healthy goal (for example a review patrol) can cross "PRs per day" or "pushes after open" without drifting.
- Which issues count. Agree: only issues the goal's sessions opened or closed.
- Bot proposes a higher limit. Agree: no proposal tool. The stop report saying what is spent is enough.
- Hours. Agree: wall clock minus held time.
- No done check. Agree: run as today and show "No done check".
- Credits before #35488. Agree: wait.
- Plan concurrency as a user-set cap. I lean no. The plan is text the Bot writes, and one go-ahead covers every round, so treating its number as "set by the user" blurs the owner-only rule §4.1 builds on. §4.3 already gives the owner-only bound: prefill
workers_at_oncefrom the plan's number and let the owner confirm it on the Goals tab. That needs no amendment to §7.2, and #37652 could then write that confirmed value instead.
A Kiro Crew team member must sign off on this design before it merges; this review does not accept it.
AI-assisted, sent by Bolin Review bot
Part of #37361
Problem / Motivation
Goal: agree a design, before code lands, that lets a Bot's owner bound one goal: count-based drift signals that stop new work, owner-set limits the Bot cannot raise, and a done check the owner approves and the Bot cannot change.
In #37361 one Bot with one goal ran for six days: 175 merge requests (99 in one day), 249 pushes after open, 16 workers at once against a plan of two or three, 62 review re-checks, 9 duplicate closes, and open issues going from 46 to 108. A maintainer triaged it as a design-level change and pointed it at the RFC process. On main, nothing limits or watches a goal:
concurrency_capis written by the Bot's ownwork_ledger_record action=goal, the goal's item total has no cap, and the patrol has no exit tied to the goal being done.Why it matters
A Bot left to run unattended can create work for itself faster than it closes it, and the owner cannot see where it stands without reading the Bot's chat. Every signal in #37361 was a count that could be read by day two.
Not a goal
accept_eval,rfc-conductor-feature-acceptance.md).What changed (motivation → approach → change)
docs/request-for-change/rfc-goal-limits-and-done-check.md: new RFC, statusdraft, tracking A Bot given a goal does not converge: it creates work for itself faster than it closes it, and the owner cannot tell where it stands #37361 and Count a Bot worker's children on other templates toward the Bot's worker cap #37538. It proposes:createwithgoal_signal_fired, and the Bot reports which signal fired and which of its own rules led there;goal_limit_reached;It cites, and does not contradict,
rfc-conductor-drift-check.md§2 and A.2,rfc-drop-goal-item-cap.md§2,rfc-work-ledger-concurrency-cap.md§7.2,rfc-crewmate-goals.md§3.1-§3.2,rfc-pr-goal-freeze.md§1 andrfc-goal-loop-approval-hold.md§3.1. Section 2.2 maps the twelve places on main where a limit or an exit is missing, read at2041e8681c. Section 5 lists feat(mate): let the owner cap how many workers a Bot keeps open #37381, Count a Bot worker's children on other templates toward the Bot's worker cap #37538 and four branches from A Bot given a goal does not converge: it creates work for itself faster than it closes it, and the owner cannot tell where it stands #37361 as open or proposal-adjacent, not shipped. Section 10 has six open decisions, each with a recommendation.docs/request-for-change/README.md: one index row at the top of the table, statusdraft.Backwards compatibility
Compatible: docs only; no behavior changes.
Removes nothing: one new Markdown file and one new index row.
Tests
Docs only; there is no code to test.
bash scripts/docs-lint.sh: all documentation checks passed;python3 scripts/docs_lint.py --strict-identifiersreports no finding in either changed file.BRAND_BASE_REF=origin/main python3 scripts/check_brand_name.py: no misspellings in the added lines.scripts/check_comment_history.pyscanssrc/kiro_crewandtestonly; this diff touches neither.scripts/verify_no_internal_host.py, run over both files: no hits.2041e8681c.Manual verification
N/A: documentation only.
Screenshots / video
Why no screenshot: no markup change; Markdown files only.
Related Issues
Part of #37361. Refs #37538, #12693, #6338, #37381, #35488.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)