# Anchor

> Rewrite a rule so anyone applies it the same way. Use when a stated rule or claim takes the reader's judgment to apply.

- Skill: `majiayu000/anchor` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/anchor`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/anchor/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/anchor

---


## Extract

From the conversation/context/user:

- The **rule** as stated
- Its **grounds** — the cases it came from, where it fires, what it decided.

## Gate

Proceed only when: A rule is already stated and the complaint is checkability "two readers could disagree where it fires".

Anything else, say which in one plain line and stop — never anchor anyway:

- Nothing stated yet → there is no rule to anchor.
- The complaint is coverage, not checkability → use **widen** skill.
- The cases changed, or the rule now serves a different objective → stop, the rule needs deriving fresh, not anchoring.

## Loop

1. Find where the judgment hides: the word or clause two readers would apply differently.
2. Replace it with something mechanical:
   - a question with a yes/no answer
   - a check a machine could run
   - a fixture with win conditions anyone reads off the same way.

   Prefer the trigger a machine could evaluate over the one a careful reader could.

3. Silently AUDIT the rewritten rule:
   it must decide every case the original decided, the same way.
   Each case it loses means you narrowed — back off by the minimum until every case fits again.

## Output

Hand back the rule with its mechanical trigger, and show it deciding one case from the grounds so the user watches it run.