Turns a red build's errors-ai.log into a pull-request comment and a job summary, so a reviewer — or a PR agent — reads the root cause instead of scrolling raw CI logs.

- run: mvn -B verify
- if: always()          # the reports only exist when something failed
  uses: stacktale/stacktale/.github/actions/report@v1.1.0

Pin to a tag; @main also works if you want the latest.

That is the whole setup. It needs pull-requests: write on the job to comment:

permissions:
  contents: read
  pull-requests: write

Output: count — the number of complete reports found, 0 when the log is absent. Useful for gating later steps.

• One comment per PR, edited in place. A new comment on every push turns a long-running branch into a wall of stale reports.
• Truncated blocks are dropped. The writer appends the header, the session marker and each block separately, so a read can land mid-write. FORMAT.md requires a reader to discard a block whose END line has not arrived rather than show half a report.
• A missing log is not a failure. A green build has nothing to say, and the action exits quietly with count=0.
• No runtime dependency for text logs. The st/1 extractor is awk, so it works on any runner. st-json/1 logs (format=json) are summarized with jq when it is on the PATH (GitHub-hosted runners include it). Without jq, the action prints a clear notice and still uploads the log as an artifact.

The library writes them. A failing test only produces one if stacktale-junit is on the test classpath — without it, a red mvn test writes nothing and this action has nothing to show. See the quickstart.

Projects configured with format=json write st-json/1 NDJSON (one JSON object per line) instead of the default st/1 text blocks. The action summarizes both when jq is available.