[Research] Set up automatic failure clustering and improvement suggestion

Summary

utils/analysis/laminar_signals/analyze.py already turns Laminar signal data into grounded recommendations. We also need it to analyze critic-failure datasets so skill improvements can be based on failed trajectories, not only on signal_events.

Target repo

• OpenHands/extensions

Dependencies

• Merge after a critic-failure dataset is available for historical conversations.

Scope

Add a second input mode for critic-failure data and keep the existing Laminar signal mode intact.

Files to update

• utils/analysis/laminar_signals/analyze.py
• utils/analysis/laminar_signals/README.md
• utils/analysis/laminar_signals/templates/default.j2
• utils/analysis/laminar_signals/templates/pr_review.j2

Acceptance criteria

• The analyzer can read a critic-failure dataset as input.
• Output stays structured and points to concrete skill or plugin files.
• Existing Laminar signal analysis still works.
• The README documents both input modes.

References

• OpenHands/extensions/utils/analysis/laminar_signals/analyze.py
• OpenHands/extensions/utils/analysis/laminar_signals/README.md

This issue was drafted by an AI assistant (OpenHands) on behalf of the user.

Start with utils/analysis/laminar_signals/analyze.py and the existing Laminar signal flow, then read README.md and both templates to understand the current structured output. The critic-failure dataset dependency must be available before this can be completed. Done means both input modes work, existing signal analysis remains intact, outputs identify concrete skill or plugin files, and the README documents both modes.