Repository navigation
docs(rfc): mark crewmate harness follows Settings partial - #33746
Conversation
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 is a docs-only status flip, the phase audit is spelled out, and it removes no readers. Notes
[DESIGN-REVIEWED] 2210fe0 Description read[DESCRIPTION-READ] d3980598177f692110f4a9300399b9e7401566f6cac12cd84d55f7ef0c1aa17c 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 |
Opus 5.5 Review — ✅ no blocking findingsReviewed Review detailsDocs-only change, no candidates. Nothing to ground. No findings. [OPUS-REVIEWED] 2210fe0 Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
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: |
bolichen97
left a comment
There was a problem hiding this comment.
Reviewed at 61fff50. This PR records an RFC decision, so I'm leaving a brief rather than an approval; the owner signs off on it.
What it does: It flips rfc-crewmate-harness-follows-settings.md from draft to accepted (2026-11-09) in both the front matter and the README index row. It fills doc-pr: 29442 and implementation-prs with the six merged code PRs. It also records one new owner decision: single-tool switches on a harness that can't honour them stay warning-only, with no gateway-side filtering and no blocked toggle for now.
Pros: The status now matches what has merged. The First Principles lane reads the RFC's status from the base commit, so this stops it from treating the merged harness work as undecided.
Cons / check: The warning-only rule is a new decision, not just a status flip. It should match what the owner actually said on 2026-11-09. implementation-prs leaves out #30116, which is correct because that PR revised the doc, not the code.
Tech debt: None added. The warning-only rule is a deliberate "for now", so a later RFC amendment will be needed if filtering is ever wanted.
CI: PR Hygiene failed at 06:22 and passed on the same head at 06:32. All other checks are green. Docs only, no code and no regex.
61fff50 to
42debaa
Compare
The owner accepted the RFC on 2026-11-09. Set status to partial and name which section 7 phases are on main and which are open, record the merged implementation PRs, and record the owner's decision that single-tool switches on harnesses that cannot honour them stay warning-only. Refs #21910
42debaa to
2210fe0
Compare
KiroCrewCaptain
left a comment
There was a problem hiding this comment.
I checked this head against current main. The audit sha 0cc173d74a is an ancestor of main. All eight cited PRs (#29442, #30116, #29509, #29498, #29490, #29495, #32849, #33540) are merged. The open items are still open: config/loader_builders.py still tells a person to set agent.member_acp_backend, and inherited_reason is absent from src/ and website/src/. The front matter (doc-pr, implementation-prs, audited-at) matches the README row. A maintainer signed off the design.
The status is accurate. The PR picks partial over accepted and names every open phase item by file. That matches the README status vocabulary.
Optional:
- The "stay warning-only" bullet (RFC line 18 and the README row) does not say which layer it covers. On main, the editor warns at save time, and session start still refuses with
capability_tool_limits_unsupported. One clause saying the start-time refusal is unchanged would stop a later PR from citing this bullet to drop it. - Sibling rows name who decided and where. Naming the owner and the place of the 2026-11-09 decision here would make it traceable the same way.
Problem / Motivation
Goal: Record that the owner accepted
docs/request-for-change/rfc-crewmate-harness-follows-settings.mdon 2026-11-09.The RFC still says
status: draft. The owner decided the direction on 2026-11-04, and the implementation PRs have merged: #29442 (RFC), #30116 (revision), #29509 (Bot harness follows Settings), #29498 (picker), #29490 (Codex worker chats), #29495 (KASdisabledTools), #32849 (layered Agent Harness page), #33540 (bell note names exact items).Why it matters
The First Principles lane reads an RFC's
statusfield from the base commit as the record of a product-shape decision. While this RFC saysdraft, later PRs inside its shape have no recorded decision to trace to, and readers of the index see a design marked as unaccepted although it shipped.Not a goal
What changed (motivation → approach → change)
Owner accepted the RFC on 2026-11-09 → README's
acceptedmeans "nothing built yet", but eight PRs have merged, so each section 7 phase was checked againstorigin/main0cc173d74a→partial, with the open phases named.status: partial,last-audited: 2026-11-09,audited-at: 0cc173d74a,doc-pr: 29442,implementation-prslists the merged code PRs.partialand names the open phases.Phase check at
0cc173d74a:kirocrew doctoradvisory row_read_backend_choice,inherited_harness_source; no doctor row for a storedagent.member_acp_backendincli_doctor.py/doctor_checks/mate_worker.py:986,dashboard/chat_turn/turn_failures.py:570,config/loader_builders.py:569still say to setagent.member_acp_backendinherited_reasonCrewEditorNextDialog.tsxhas no harness field; noinherited_reasoninsrc/crewmate_harness_notice.py,AgentBackendTab.tsxcrewmate-harnessline), #32849, #33540test/fixtures/acp_frames/codex/worker-kit-live.jsonl)kas_pin_harnesses,_note_pin_fallbackincrewmate_harness.py)judges_a_model_edit,kiro_spelling_applies); start-time drop of an invalid own model and the editor note not foundBackwards compatibility
Compatible: docs only; no code, config or contract changes.
Removes nothing: the diff edits two Markdown files' status text and takes nothing away.
Tests
Proof: not provable -- docs or config only
scripts/docs-lint.sh: all documentation checks passed.Manual verification
All eight PRs above read as
MERGEDintomainviagh pr view.Related Issues
Refs #21910
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.