01 — Start autonomous work

Format

Text Parameterized

Outputs

result (text)

Replace the ${…} placeholders with your values before using this prompt. See parameters and defaults below.

Read AGENTS.md and follow the project protocol.

Then:

1. Read PROJECT.md and WORK.md.
2. Consult DECISIONS.md and active plans only when relevant.
3. Inspect the repository’s actual state, including Git, local changes, branches, and relevant tests.
4. Identify active tasks, blockers, reserved files, and the legitimate next step.
5. Check whether there is an authorized task you can advance without a decision from me.
6. Take on or resume the most appropriate task and record your responsibility in WORK.md.
7. Classify the task as trivial, normal, or complex, and as R0, R1, R2, or R3.
8. Plan only to a depth proportional to the work.
9. Rename the session, when the platform allows, using:
   ${identificador-opcional} ${tarefa-curta} — ${atividade-atual}
10. Choose freely between working on the current branch, creating a branch, or using a worktree, according to conflict risk and project policy.
11. Autonomously carry out all actions covered by the mandate.
12. Verify the result with tests and evidence proportional to the risk.
13. Decide whether cross-review is necessary.
14. Update WORK.md and any affected canonical documents.
15. Finish using the project’s completion contract.

Do not ask me to choose reversible technical details.

Do not revisit approved decisions without new evidence.

Do not create duplicate documentation.

Do not perform restricted external actions without a mandate.

Stop only if:

- there is no authorized work;
- there is an unresolvable conflict;
- a decision falls outside the mandate;
- there is a material risk that requires my input;
- or all applicable work is complete.