| diataxis_type | how-to |
|---|---|
| title | How to Integrate with Skill Creator |
| description | Use autoresearch alongside the skill-creator plugin for end-to-end skill development |
Autoresearch depends on the skill-creator plugin and works well as part of a broader skill development workflow.
Autoresearch uses skill-creator's grader agent to evaluate skill outputs. The grader:
- Reads eval expectations and the skill's execution transcript
- Produces
grading.jsonwith pass/fail for each expectation - Provides
eval_feedbackwith suggestions for improving evals
Autoresearch locates skill-creator automatically:
${CLAUDE_CONFIG_DIR:-${HOME}/.claude}/plugins/cache/claude-plugins-official/skill-creator/*/skills/skill-creator/If not found, autoresearch tells you to install it.
After a successful autoresearch run, the skill's behavior may have changed significantly. The trigger description in SKILL.md's frontmatter may no longer accurately describe what the skill does.
Use skill-creator to re-optimize the description:
/skill-creator optimize-description path/to/my-skillThis analyzes the updated SKILL.md and rewrites the description field in the frontmatter to better match the skill's current capabilities.
- Create the skill:
/skill-creator path/to/my-skill - Create evals:
/autoresearch --eval-doctor path/to/my-skill - Improve the skill:
/autoresearch path/to/my-skill - Apply changes: Confirm when prompted
- Optimize description:
/skill-creator optimize-description path/to/my-skill
- Check evals: Verify
evals/evals.jsonexists and is adequate - Run improvement loop:
/autoresearch path/to/my-skill - Review and apply: Check convergence report, apply if satisfied
- Update description if needed:
/skill-creator optimize-description path/to/my-skill
Autoresearch uses the same eval schemas as skill-creator. See Eval Schema for the evals.json and trigger-eval.json formats.
The grading.json output schema is defined in skill-creator's references/schemas.md.