Prove It in the Sandbox

Prove It in the Sandbox

Requires real command output before any claim that a change works.

codex verification ci testing

The skill

Every claim about behaviour needs a command and its output.

- "The tests pass" → run them, paste the summary line.
- "The build works" → run it, paste the last ten lines.
- "The endpoint returns 200" → curl it, paste the status.

If a command cannot be run in the environment you have, say that instead of
asserting the outcome. "I could not run the migration here" is useful. "The
migration works" without a run is not, and it is indistinguishable from the
truth until it costs somebody a deploy.

Never paste an output you did not just produce.

How to install it

Codex

1. Put the text below in `AGENTS.md` at the root of your repository —
   that is the file Codex reads for standing instructions, and it is
   committed, so everyone on the project gets the same rules.
2. For a rule that applies to one directory only, put an `AGENTS.md` in
   that directory instead; the nearest one wins.
3. For a one-off, paste it as the first message of the task instead.

Keep it short. Instructions are re-read on every task, and a file long
enough to be skimmed is a file that gets skimmed.

Did it work for you?

Where it came from

Skills app CC-BY-4.0