Human-Agent Responsibility Matrix

Procedure

1. 1. Map each business task to references/task-contract.md without credentials or sensitive payloads.
2. Run scripts/build_responsibility_matrix.py with an approved governance policy, previous state, and fixed review date.
3. Resolve every blocked task, then have the named owner and approver review the generated deployment gate before enabling automation.

Guardrails

• - The workflow audits proposed responsibilities; it never executes the inventoried actions.
• Agents cannot approve their own actions or change the governance policy.
• External sends, writes, deletes, payments, approvals, regulated data, and irreversible actions require explicit human control.
• Preserve owner, reviewer, approver, expected evidence, approval SLA, rollback plan, and policy reason.
• Use next_state.json to distinguish changed task definitions from already-reviewed versions.

Outputs

The workflow creates a responsibility matrix, gap register, approval queue, machine-readable approval policy, RACI report, deployment gate, audit log, and next-run state.