challenge-gen

To generate training challenges from agent diagnostics, you use the challenge-gen Skill with an agent name after running a profile diagnosis. It processes the agent struggle profile to produce targeted edge-case practice and commit-replay tasks for code-focused agents.

Commit-replay training for code-focused agents is a practice method that uses real codebase commit history to create realistic edge-case scenarios. It turns diagnosed weak spots into targeted, repeatable practice tasks instead of vague improvement goals.

You create grounded edge-case scenarios by feeding an agent struggle profile or basic profiling input into a challenge generator. The system pulls from real codebase edge cases and commit history to package scenarios with acceptance criteria and hidden traps.

Yes, running a diagnosis before generating practice challenges is required to identify weaknesses, gaps, and calibration needs. The challenge generation workflow validates agent ownership and uses the diagnostic output to select grounded scenarios for targeted improvement.

Yes, you can use basic profiling input instead of full diagnostics for challenge generation. The system accepts both detailed diagnostic output and basic profiling data to identify agent weaknesses and design calibrated edge-case training scenarios.

The best way to structure active-learning scenarios for agent improvement is to package them as a structured challenge set. This includes specific scenarios, acceptance criteria, hidden traps, and provenance metadata to support downstream learning workflows.