Repository-native guidance for coding-agent work.

A good task definition states the objective, implementation boundary, acceptance criteria, and executable validation before the agent changes code.

Declare allowed target paths before implementation, review changed paths against that boundary, and require escalation before the agent broadens scope.

Put executable validation in the task contract, require the observed result in the implementation summary, and treat missing evidence as incomplete work.

Review the declared boundary, changed paths, validation evidence, and risk areas first, then inspect code where the evidence or scope requires it.

Require a written implementation summary with changed paths, validation results, unresolved risks, and an acceptance or follow-up decision.

A concise checklist for handing an implementation task to a coding agent and receiving work back with usable evidence.

Store durable intent, scope, validation, decisions, and implementation evidence in repository artifacts instead of relying on remembered chat context.

A Human-Agent-Contract is a chain of versioned repository artifacts from intent through validation evidence and acceptance.

Trace the specification through planning, then create tasks that have a single verifiable outcome, bounded paths, and explicit validation.

Start with one bounded, reversible task and establish repository-visible validation and review evidence before expanding use.

Use coding assistants for implementation; keep the durable implementation contract and evidence in versioned repository files.

Keep planning and coordination in your tracker; govern the implementation contract and validation evidence in the repository.