8 skills build & grade Claude Code skills/agents. Create, judge, eval, ship.
Eight skills covering the full lifecycle of an AI skill or agent definition: create it, judge it, review it before it ships, apply approved findings, evaluate it behaviorally, recap what it actually does versus what it claims, update it, and audit a whole repo of them at once.
These skills follow the Agent Skills specification so they can be used by any skills-compatible agent.
npx skills add robcsaszar/ai-forge
/plugin marketplace add robcsaszar/ai-forge
/plugin install robcsaszar-ai-forge@ai-forge
Copy the skills/ directory into your project's .claude/skills/.
npx skills add robcsaszar/ai-forge --skill ai-forge-judge
Or manually, copy just that skill's directory:
cp -r skills/ai-forge-judge /path/to/project/.claude/skills/| Skill | Description |
|---|---|
| ai-forge-apply | HITL application of a numbered list one item at a time: status board upfront, per-item approve/skip, commits after each approved item |
| ai-forge-audit | Batch-evaluates every skill and agent in a repo with ai-forge-judge and renders one consolidated grade report, worst first |
| ai-forge-create | Creates a new skill, agent definition, or instruction file via discovery recap, pattern selection, and a judge + apply quality gate |
| ai-forge-eval | Behavioral eval for skills and agents: parallel with-artifact vs baseline runs, assertion grading, blind A/B comparison, and optional benchmark trend tracking |
| ai-forge-judge | Evaluates any LLM prompt for quality: grouped dimensional scoring, letter grade, and a step-through-ready improvements list |
| ai-forge-recap | Reads a skill or agent's actual body and reports what it really does versus what its description claims |
| ai-forge-review | Critically stress-tests an agent, skill, or AI workflow definition before it ships |
| ai-forge-update | Updates an existing skill or agent: structured recap, drift detection, change elicitation, and a post-change quality gate |
ai-forge-create and ai-forge-eval ship small helper scripts. Read SAFETY.md for what each does.
MIT © Rob Csaszar