A thin lens between your message and the agent's action.
Most AI agents rush to act — burning context and tokens on misunderstood requests.
Clarify pauses just long enough to detect ambiguity, then asks at most one question before the agent touches any code.
CLEAR REQUEST? → Act immediately. No ceremony.
AMBIGUOUS? → Ask exactly one question. Then wait.
Seven signals of ambiguity — a quick internal scan before every action:
| # | Signal | Trigger Example |
|---|---|---|
| 1 | Unstated scope | "add auth" — which method? which routes? |
| 2 | Missing constraints | "make it faster" — by how much? at what cost? |
| 3 | Vague criteria | "make it better" — by whose standard? |
| 4 | Assumed context | references to files not yet established |
| 5 | Multiple readings | "refactor the auth" — extract? rename? restructure? |
| 6 | Unstated trade-offs | "use a database" — which one? why? |
| 7 | Emotional requests | "this feels messy" — where? what specifically? |
No ambiguity detected? Zero overhead — the agent proceeds immediately.
npx add-skill ccjh11/clarifyWorks with Claude Code · Codex CLI · Cursor · Reasonix and any agent following the Agent Skills spec.
Clarify is not a workflow. It's not brainstorming, not planning, not a design phase.
- One question max. Multi-step interviews belong elsewhere. Pick the most consequential ambiguity.
- Skip freely. Follow-ups, test commands, typo fixes — don't overthink.
- A lens, not a gate. It sharpens intent. It does not block.
License · MIT · ccjh11