Skip to content

Triage: a silent step's name, script and exit line reach the judge; an under-floor red is said as that - #639

Merged
kirkmarple-clockgrove merged 2 commits into
mainfrom
fix/637
Nov 19, 2026
Merged

kirkmarple-clockgrove merged 2 commits into
mainfrom
fix/637

Conversation

@kirkmarple-clockgrove

Copy link
Copy Markdown
Contributor

Closes #637

What changes

1. A failed step with no failure line gives the judge evidence. When a failing job's log yields no failure signature, the triage question adds the failed step's name (from the job's steps, already read), the run: script the runner echoed in the log's ##[group]Run … block (cut at the 4,000-character failure window cap) and its exit line, between random-tag markers as data. No new GitHub read. A step that runs an action has no script and says so. Questions for jobs with a failure line are unchanged.

2. A red that is the change's only by default says so. When triage finds no flake, no job's answer decides the red, and a verdict is under the floor, classify returns that (unsure). The queue line, the issue's queue-red record line, the fix pass's ask and the Block then read, for example:

Pendulum could not tell whose failure it is (base_failure 0.64, under the 0.7 floor; no failure line in the log), so it treated it as the change's

and never "the PR's own change". The triage line's outcome reads "no verdict at the floor, so treated as the change's: fix pass". The stored queue-red record, its kind: own, the Block cause code and the markers are unchanged; the wording is never stored.

Tests

Two, each failing on main:

  • plugin/test/triage.test.mjs: a silent step's question carries its name, script and exit line between the markers (through openTriage().classify).
  • plugin/test/queue-watch.test.mjs: an under-floor queue red's fixing line, Block line and Block reason have no "own change" and carry the verdict, score and floor.

Not in this PR

Validation

Lint; the plugin and bench suites, each with and without GITHUB_GRAPHQL_URL. One rerun: cross-os.test.mjs #518 B1 failed once in the plugin run with GITHUB_GRAPHQL_URL and passed on the rerun.

🤖 Generated with Claude Code

https://claude.ai/code/session_017VqP6hg8ahWzteM9zqi8pU

…-floor red as that (#637)

A failed step that printed no failure line now adds its name, the run
script the runner echoed in the log and its exit line to the triage
question, between random-tag markers as data.

A red merge-queue run that is the change's only because no triage verdict
reached the floor says so on the queue line, the issue's record line, the
fix pass's ask and the Block, with the best verdict, its score and the
floor, and never as the PR's own change. The stored records and cause
codes are unchanged.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017VqP6hg8ahWzteM9zqi8pU
…637 review)

A base_failure at or above the floor that the default branch did not confirm keeps today's wording; the mixed case no longer drops it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017VqP6hg8ahWzteM9zqi8pU
@kirkmarple-clockgrove
kirkmarple-clockgrove added this pull request to the merge queue Nov 19, 2026
Merged via the queue into main with commit d9d7fef Nov 19, 2026
10 checks passed
@kirkmarple-clockgrove
kirkmarple-clockgrove deleted the fix/637 branch November 19, 2026 00:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Queue triage: a silent step failure gives the judge no evidence, and an under-floor red is reported as 'this PR's own change'

1 participant