Problem: When claude-code-action produces a low-quality run (partial output, truncated response, or a run that touches 0 files when it should have made changes), there's no mechanism to detect this and retry before the workflow completes.
Currently users have to manually review every run or build custom post-processing scripts to catch these cases.
Proposal: Add optional quality thresholds that trigger automatic retry:
\\yaml
- uses: anthropics/claude-code-action@v1
with:
quality_checks:
min_files_changed: 1
max_retries: 2
fail_on_empty_diff: true
fail_on_truncated: true
\\
Why this matters:
This is especially relevant for automation-triggered runs (cron, issue assignment) where no human is watching in real-time.
Happy to contribute on this — I run 16 automated agent jobs daily and have built retry/quality logic for exactly this use case.
Problem: When claude-code-action produces a low-quality run (partial output, truncated response, or a run that touches 0 files when it should have made changes), there's no mechanism to detect this and retry before the workflow completes.
Currently users have to manually review every run or build custom post-processing scripts to catch these cases.
Proposal: Add optional quality thresholds that trigger automatic retry:
\\yaml
with:
quality_checks:
min_files_changed: 1
max_retries: 2
fail_on_empty_diff: true
fail_on_truncated: true
\\
Why this matters:
This is especially relevant for automation-triggered runs (cron, issue assignment) where no human is watching in real-time.
Happy to contribute on this — I run 16 automated agent jobs daily and have built retry/quality logic for exactly this use case.