Cold-eye is a readiness pass on anything shipped, deemed done, or ready to ship. An agent reads the skill and writes a critique. Verdict first, then a ranked list. The subject does not change.

Get started: pick the agent, install the skill, then run the sample checklist at coldeye.dev/docs/install.

Claude.ai without project files is not a supported first-use route. The install page lists current host limits.

Site: coldeye.dev

One example (illustrative)

You wrote: “Install the package and launch the app.”

Cold-eye finds: the guide names the package but never gives the launch command. A new user cannot finish setup from the instructions given.

You get a ranked finding in the critique file:

**Verdict:** close

1. **F-001** · test 2 · invented
   > Install the package and launch the app.
   Cold reader: installs the package, then guesses a launch command or stops.
   Put: The exact launch command on the next line, and what the reader sees when the app is running.

The guide itself is unchanged.

First run

Save until-ready.md (or the checklist below), then ask:

Use Cold-eye on until-ready.md. Follow the installed Cold-eye skill. Write the critique. Leave the checklist unchanged.

# Review until ready

Fictional procedure for desk-stamp notes. Labeled example.

1. Open `notes.md`.
2. Read every section.
3. Write findings next to the file as `notes.review.md`.
4. Repeat the review until ready.

The critique lands in until-ready.cold-eye.md. The checklist should be unchanged. A finding should point at step 4, the unresolved “until ready” condition. Wording varies by model. A critique alone does not prove the skill loaded: check the agent's skill list, or ask it to quote the first heading of SKILL.md.

A writable workspace is required. Package coldeye, skill folder cold-eye.