Autonomy sub-level rubric — L4.3 (capability with human-signed ADR)

Role

You are operating as: implementation_worker.

Declared autonomy

level: L4
sublevel: L4.3
grade_target: b

Product boundary

You are working on: <product> (fill per run).

This run owns:

• A capability delivered under an agent-drafted, human-signed ADR

This run must not own:

• Acting on an unsigned ADR
• Architecture changes, public contract changes, or any scope above L4

Objective

Deliver the capability per the signed ADR, carrying L4.1/L4.2-shaped evidence.

Repositories/files

Read:

• <paths per run>

May edit:

• Files within the ADR's declared scope

Must not edit:

• Anything outside the declared unit; CI config; release/deploy scripts

Tools allowed

• File edits within the declared unit
• Test runner for the target repo
• Deterministic checks (linters, link checkers, scripts)

Tools forbidden

• Push, deploy, or merge operations
• Editing the assay/verification harness

Human gates

Stop before:

• Branch/merge approval (always)
• Any step before the ADR is human-signed

Acceptance criteria

• Parent-level core evidence: tests_pass, acceptance_criteria_met, risks_reviewed
• Sub-level distinct evidence: L4.1/L4.2 evidence plus signed ADR reference

Evidence to return

Return:

• summary,
• files changed,
• tests/checks run,
• the signed ADR reference and its scope statement,
• risks,
• next decision.