Skill Checker Agent
The background agent that evaluates work and generates skills.
How It Works
After each prompt, Skilltrace spawns a lightweight background agent that reads the latest transcript and evaluates whether meaningful, reproducible work was completed.
The agent operates in its own isolated context and never interferes with your main conversation.
Evaluation Criteria
The skill checker looks for work that is:
- Complete:the task reached a working state
- Reproducible:steps could be replayed in a fresh environment
- Specific:concrete tools, commands, and configurations
- Non-trivial:meaningful enough to save future effort
Skill Generation
When skill-worthy work is identified, the checker generates:
- A descriptive name and one-line summary
- Step-by-step instructions with exact commands
- File paths and configuration values
- Prerequisites and environment requirements
- Project association for organization
Versioning
If you repeat a similar task with a different approach, the checker creates a new version of the existing skill rather than a duplicate. Old versions are archived, not deleted.
Agent Definition
The skill checker agent is defined in agents/skill-checker.md. It contains the evaluation rubric, skill template, and versioning logic.