skill-ab-evallisted
Install: claude install-skill cskwork/skill-ab-eval
# skill-ab-eval — prove what actually works
Two questions, one harness, real evidence:
1. **Skill axis** — does this `SKILL.md` change the agent's behavior, or is it dead
weight in the context window? (`with_skill` vs `without_skill`)
2. **Harness axis** — for *this* task or domain, which CLI agent is best?
(`claude` vs `codex` vs `gemini` vs `agy` vs `openai`)
Run them separately or crossed. It works on **any task you give** — a skill's
evals, or an ad-hoc prompt you type — across any domain.
It's the agent-native cousin of
[agent-skills-eval](https://github.com/darkrishabh/agent-skills-eval) (same
`evals.json` and with/without model) wired to the multi-CLI delegation pattern of
[cc-agent-call](https://github.com/cskwork/cc-agent-call): where cc-agent-call
*routes* work to the best CLI, this *measures* which CLI is best.
## Mental model
```
one task prompt (a skill's eval, or one you type)
│
┌────────────────────────┴────────────────────────┐
│ matrix: harness × {with_skill, without_skill} │
▼ ▼
claude with / without codex with / without ...gemini, agy, openai
│ each cell = a FRESH context (subagent or separate CLI process)
└────────────────────────┬────────────────────────┘
▼
judge (any harness)
grades each answer vs the assertions