Repository navigation
feat(claude-remote): steward fix_mode never on feat/agents- entries - #1508
Conversation
Owner decision 2026-10-03: on DevLoop feat/agents-* PRs the shepherd CWFT is the only actor that fixes review findings; the steward runs no model tick until the PR is approved at its current head and then owns only the merge path. On mctl-telegram#730 both actors fixed the same findings and the steward's tick lost a non-fast-forward race to the shepherd's push ($1.64 wasted); #715 and #722 show the same overlap. Sets "fix_mode": "never" on the four feat/agents- entries only (claude/ and fix/ keep today's steward behaviour) and corrects the ownership comment, which still described the pre-issue-292 repo partition. Needs image 0.13.3; removing the key restores today's behaviour. Claude-Session: https://claude.ai/code/session_012kWiYf1BX6wcXB2kc8x15U
|
Claude finished @mashkovd's task in 3m 22s —— View job Reviewing PR #1508
Verdict: APPROVE (no P1/P2, 1 P3) Reviewed
P3 (nit, inline): the rewritten top comment now references both Note: |
Antigravity (agy) review — informationalModel: gemini-3.8-flash-medium · commit No significant issues found. |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
No P1/P2 findings (1 P3). The fix_mode: never key is correctly added to exactly the four feat/agents- entries named in the PR description (mctl-design, mctl-gitops, mctl-telegram, mctl-pairdesk); claude/* and fix/* entries are untouched. JSON structure (commas/braces) is valid. One P3 left inline: the updated top comment references both SHEPHERD_SKIP_SERVICES and SHEPHERD_FIX_ONLY_SERVICES, but the untouched per-repo comment for mctl-pairdesk still says it's in SHEPHERD_SKIP_SERVICES — worth reconciling so a future reader doesn't conclude review findings on pairdesk feat/agents- PRs have no owner. Good to merge.
SHEPHERD_SKIP_SERVICES is only mctl-academy since issue-292; mctl-pairdesk moved to SHEPHERD_FIX_ONLY_SERVICES (cwft-mctl-agents-shepherd.yaml). Claude-Session: https://claude.ai/code/session_012kWiYf1BX6wcXB2kc8x15U
|
Claude finished @mashkovd's task in 1m 3s —— View job Reviewing PR #1508 (commit
|
Antigravity (agy) review — informationalModel: gemini-3.8-flash-medium · commit No significant issues found. |
There was a problem hiding this comment.
Config-value-only diff: fix_mode: never added to exactly the 4 named feat/agents- entries; the follow-up commit also fixes the stale mctl-pairdesk comment (confirmed against cwft-mctl-agents-shepherd.yaml: pairdesk is indeed in SHEPHERD_FIX_ONLY_SERVICES, not SKIP). JSON valid, no secrets/RBAC/resource changes.
Summary
This turns on the owner-approved single-owner rule for review remediation on DevLoop PRs. The code shipped in mctlhq/mctl-claude-remote#78 (image 0.13.3).
It adds
"fix_mode": "never"to the fourfeat/agents-steward entries:The rule for those entries:
SHEPHERD_FIX_ONLY_SERVICES.Invariant: for one PR/head in the review-remediation phase, exactly one actor makes changes.
claude/*andfix/*entries are unchanged: the steward still owns them end to end. The key is set per entry rather than at the top level for exactly that reason. The ownership comment in the file is also corrected; it still described the pre-issue-292 partition.Why
20261003T010247Zlost arace-non-fast-forwardagainst shepherd commit4db2345. Both actors were fixing the same findings, and the losing tick cost $1.64.Rollout and rollback
labs-claude-remoteruns 0.13.3. Older images ignore the key, so merging early is harmless, but merging after keeps the effect aligned with the image.steward.logshows noaddress-reviewfrom the steward, every non-human fix push comes from mctl-agents[bot], and the merge still lands assteward:merged.auto, which is today's behaviour.🤖 Generated with Claude Code
https://claude.ai/code/session_012kWiYf1BX6wcXB2kc8x15U