Discussion-summary validator fails the run when a finding thread describes a failed check #722

Summary

A follow-up review fails before any reviewer runs when an existing finding thread is about CI check behavior. The discussion-summary step summarizes the thread faithfully, the summary contains a phrase such as "failed check", and validateDiscussionSummaryText rejects it as reviewer-facing process state. After one retry the whole run exits with:

structured output invalid after retry: discussion summary text contains excluded reviewer-facing process state

Nothing is posted, so the earlier review stays in place and the pull request cannot get a fresh verdict.

Reproduction

1. Open a pull request that changes a CI workflow.
2. Let cr post an inline finding whose subject is check behavior, for example that a skipped job lets a failed check pass.
3. Push a fix and run cr review (or cr review --rerun).

Both attempts of the discussion-summary task return summaries like "…allowing a failed check to pass without validation…". The phrase matches \bfailed (?:build|builds|check|checks)\b in forbiddenDiscussionSummaryPatterns (internal/dossier/dossier.go), so every follow-up run on that pull request fails the same way.

Seen on 0.10.367. The patterns and validation are unchanged on main (v0.10.371).

Expected

A summary of a finding about how checks behave is not a claim about this pull request's CI status, and should not fail the run. #559 fixed the same class of false positive for "approval".

Possible directions

• Narrow the patterns to assertions about this pull request's own state rather than any mention of a failed check.
• When one thread summary is rejected, drop or fall back for that entry instead of failing the whole invocation (related: One failed thread analysis fails the invocation instead of degrading; thread tasks never use the isolated failure status #604).