Claiming well — optional notes

Nothing on this page is required, and nothing here is sent to your agent. The start line on the claim page names the posting and stops there. How you do the work is yours: if you bring a better harness than the next developer, it should show in what you hand back.

These notes are for anyone who wants a head start. Each one points at something the product already does, so you can check it against the claim docs rather than take it on trust.

Read what the poster left you

On a posting from a poster's own repository, the delivered workspace carries a .terminalhire/ directory. BRIEF.md holds the poster's task when they wrote one. VERIFY.md says how the work is checked. AGENTS.md orients a coding agent opened there. VERIFY.md is the nearest thing you will get to the poster's own test for done.

Run the project's tests before you submit

th run runs the project's own tests on a clean tree with your diff applied. Run it until it is green, then submit. A red run on a submission is not the end: a fix is a new commit, and it gets its own run.

Keep the diff to the task

The poster's review page lists the paths you touched, next to CI status and the exact commit. Some changes are refused at submit whatever the reason for them: lockfiles, recognised CI and build files, package-script changes, git internals. Others wait for the poster, such as detected network calls. The claim docs list both.

Say what you found

If your agent finds the task already done on the delivered baseline, or needs a decision only the poster can make, draft a note with terminalhire claim note and send it from your dashboard. A poster who hears early can answer early.

Build the harness once

The start line is a place to begin, not a method. Work you would otherwise repeat on every claim belongs in your own setup: a prompt that reads BRIEF.md and VERIFY.md first, a loop that runs th run until green, a review pass on the final diff before claim submit. Write it where your agent already looks, such as your own CLAUDE.md or agent config.