Lane telemetry: babysit-loop

[kyle-sexton]

Per-lane telemetry tracker for the babysit merge lane (/source-control:babysit-loop). Each cycle upserts one sentinel-identified comment below, keyed by lane instance ID.

This issue is lane infrastructure: it is never worked as backlog, never gates a drain exit, and is excluded from attention views.

Supersedes #232, which was closed on the 2026-08-12 drain with the note that "future lane cycles can open a fresh tracker if needed."

_{Opened by an AI agent (source-control:babysit-loop coordinator, instance ccr-session-babysit-loop-20260905).}

[kyle-sexton]

Babysit lane — cycle 1 (instance ccr-session-babysit-loop-20260905, lane=babysit-loop)

Tier: autopilot (babysit_autopilot_merge_tier unset → disabled, so merge authority is the base pinned gate). Scope: single PR, #552. Merged 0, advanced 0, held-by-policy 1, blocked-by-defect 0, yielded-to-foreign 0.

PR #552 — anthropics/claude-code-action 1.0.215 → 1.0.216 (dependabot)

Head 8df63bd430273b21c41a4133614df002b3f70ffe, unchanged across the whole cycle. mergeable_state clean, combined status ci-lanes success, 30/30 lane checks green, no reviews, no unresolved threads. Nothing was committed, pushed, resolved, or merged.

The needs-issue-linkage label was a false blocker. The scout snapshot flagged a pr-contract:linkage bot comment listing the four missing body sections and the missing closing keyword. The required gate is the pr-issue-linkage / pr-issue-linkage check, and it is success on this head — the bot comment is advisory output, not the gate. do-not-merge / do-not-merge is also success, so the label gates nothing mechanically. No PR-body edit was made: the contract is already satisfied as the check scores it, and a body edit on a Dependabot PR is overwritten by ·@·d·ependabot r·ecreate anyway.

One apparent red check was a superseded run, not a failure. ci-status reports failure on run 33931491511, whose every sibling job is cancelled or skipped; the live run 33931491620 reports ci-status success at 00:01:37Z. No re-run was fired — nothing here reads as flaky.

Held, for two independent reasons:

1. Dependency hold-merge. babysit_merge.py refuses a dependency-manager-authored PR in every tier absent --allow-dependency, a human decision never passed autonomously (babysit-prs/reference/safety.md, Guarded Mutation Wrappers; and Stop And Ask, "dependency acceptance is human judgment in every tier"). Consistent with the cycle-23 finding on build: bump the github-actions group across 3 directories with 3 updates #281 that the gate keys on PR author, not head-commit author.
2. The merge gate could not run, so ready is unproven, never inferred from the green rollup.

ESCALATION: merge gate is unrunnable on this runner — gh predates closingIssuesReferences

The sanctioned read-only form

bash "${CLAUDE_PLUGIN_ROOT}/bin/source-control-babysit-merge" melodic-software/ci-workflows#552 --allowed-owners melodic-software --self-logins @me

returned a script-level error, not a harness denial and not a gate verdict:

gh pr view 552 -R melodic-software/ci-workflows --json state,…,closingIssuesReferences failed:
Unknown JSON field: "closingIssuesReferences"

The runner carries gh 2.45.0; closingIssuesReferences is not among the fields that version exposes on pr view. babysit_merge.py:946 requests it unconditionally, in the single gh pr view call the whole gate depends on, even though the value is only consumed at line 1089 for the autopilot-merge-tier issue-linked criterion — a tier that ships disabled. So a field needed by a disabled tier takes the entire merge gate down on any runner below that gh floor.

This is distinct from the cycle-23 escalation (a harness auto-mode classifier denial of the wrapper's command form, later found not to reproduce). This one is deterministic and environmental: it will hit every PR on this runner, not just dependency PRs, until either the runner's gh is upgraded or the field is requested conditionally / tolerated as absent. Per safety.md the call was not retried and no fallback path was used to synthesize a verdict.

Suggested fix, for a human to rule on: make the closingIssuesReferences field request conditional on --autopilot-merge-tier, or degrade gracefully when gh rejects it, so the base gate keeps working on older gh.

Human floor — what #552 needs

Dependency acceptance of a patch bump whose sole upstream commit is d75b94d ("bump Claude Code to 2.1.261 and Agent SDK to 0.3.261"). Green CI is not acceptance. Either merge it directly or instruct a lane run to pass --allow-dependency for this specific PR.

Hygiene

Lane-claim marker posted and released on #552 (claim, release). No plugin worker lease was acquired — no PR-specific mutation was performed, and the coordinator's marker comments are a separate mechanism from the plugin's worker lease. No worktree created, no branch pushed, no thread resolved, no settings touched. No foreign lane claim or foreign activity was present at claim time or at release.

{"schema":"source-control/babysit-loop-state@1","cycle":1,"backoff_level":0,"stop_mode":"drain","tier":"autopilot","merge_rung":"base-pinned-gate","rate_limit_latch":false,"guard_mode":"proactive","instance":"ccr-session-babysit-loop-20260905","merge_gate":"unproven-gh-field-unsupported","restart_request":null}

---

Generated by Claude Code