Unnamed repository; edit this file 'description' to name the repository. http://cgit.235523.xyz/tinyusb.git/atom/?h=claude%2Fvalidator-done-signals 2026-08-27T17:37:29Z 2026-08-27T17:37:29Z hathach thach@tinyusb.org 2026-08-27T17:37:29Z urn:sha1:9c521da79fef57e30d806d3d32d777268d7a9118 Committer date can precede the actual push by hours, making a leftover error/quota comment from an earlier push look fresh; use the head SHA's check-suite creation time instead, falling back to committer date only when no check suites exist. 2026-08-27T17:20:46Z hathach thach@tinyusb.org 2026-08-27T17:20:46Z urn:sha1:9480bf4e0d684046c46caeb1d02635c7188db21b Codex's "Something went wrong" comment carries no Reviewed-commit line, so a stale one from an earlier push was being read as settled; gate it on created_at postdating the head commit instead. The unfiltered check-runs listing is paginated and can drop claude-review off a PR with many checks — fetch it by check_name instead. 2026-08-27T17:10:31Z hathach thach@tinyusb.org 2026-08-27T17:10:31Z urn:sha1:399fa9616a92986d629c782635159a2e9ae40dea Codex's issue comment now must match the head SHA via its Reviewed-commit line before counting as settled, so a stale verdict from an earlier push isn't mistaken for done. pr-babysit no longer sleeps out the last cycle when auto-review is still pending, since nothing re-checks after it. 2026-08-27T09:08:45Z hathach thach@tinyusb.org 2026-08-27T09:08:45Z urn:sha1:3b9c52003659d1495bc2db97b051bfa8c6157612 Check PR reviews' commit_id and the Claude check run's completed status against headRefOid instead of any review comment's presence, so a stale review from a prior push no longer reads as settled. pr-babysit now backs off with an increasing wait when no bot has reported yet instead of re-harvesting the same unchanged PR every cycle. 2026-08-27T08:49:42Z hathach thach@tinyusb.org 2026-08-27T08:49:42Z urn:sha1:b6fb315c9ffd87531a94609500a6b9c6d0292c2b done=true on zero findings let a cycle that ran before the bots posted read as complete; with a fast-green CI the babysit loop could exit unreviewed. Settled = verdict posted (Copilot review header, Codex verdict comment, Claude review or concluded check) or a reported usage/quota limit, so a rate-limited bot cannot hold the loop hostage. 2026-08-27T08:09:49Z Ha Thach thach@tinyusb.org 2026-08-27T08:09:49Z urn:sha1:42db982e7c6e8ee7d2f80d15a93a7092d3b708c9 test/hil: drop the Windows accommodations, which accommodate nothing 2026-08-27T07:56:23Z Ha Thach thach@tinyusb.org 2026-08-27T07:56:23Z urn:sha1:677182694f555c451eaec0a0d4ee7b14ac532783 workflows/agents: overlap pr-babysit's review and CI lanes; split pr-monitor; pin agent efforts 2026-08-27T05:21:04Z hathach thach@tinyusb.org 2026-08-27T05:21:04Z urn:sha1:92b38fc3b5e9080c96294079e6117d74429f3448 The claude stage reviews the diff directly (the code-review skill is a CLI built-in, unavailable to subagents); the gate is enforced in-script from structured findings, failing only on confirmed correctness/safety bugs. 2026-08-27T05:21:04Z hathach thach@tinyusb.org 2026-08-27T05:21:04Z urn:sha1:208f82efe8b9dc7ca3eb210dcbf455a14c97cf3c Review findings are validated, fixed, and pushed without waiting on CI; checkoutDir decouples the PR checkout from the session cwd. File-less CI failures are scoped by a dedicated agent, paths canonicalized and existence-checked via git ls-files, overlapping groups merged. Per-id reply/resolve accounting retries failures and holds the green exit until all outward work is drained. 2026-08-27T05:21:04Z hathach thach@tinyusb.org 2026-08-27T05:21:04Z urn:sha1:2b9ddd8b00eb57225fd9caeecd0da1a7954aee06