structured-autonomy-plan

Restate the goal in one sentence, decompose it into atomic tasks with explicit files and validation criteria, then sequence tasks by declared dependencies. Finish with a mandatory test strategy covering unit, integration, and acceptance criteria.

Each task must be completable in isolation, verifiable with a clear done signal, and bounded in scope. Dependencies must be explicit via depends_on fields, and the plan must include enough detail for cold-start execution without clarification.

Use it after architect approval, for work requiring more than three discrete changes, for cross-agent handoffs, or for test-driven work. Single-step changes do not need a formal plan.

Avoid vague tasks like implement the feature, missing test strategies, implicit dependencies, and unbounded tasks touching more than five files. Each task needs explicit files, validation, and declared dependencies.

The plan includes an entropy estimate counting files to create and modify, classified into GREEN, YELLOW, or RED zones. Risks are listed with mitigations and severity ratings so reviewers can gate execution.