gov-autonomous-execution-policy

What problem does it solve?

Many projects lack clear, enforceable rules for when AI agents may run without human intervention, which risks unauthorized changes, missing audit trails, and insecure sandbox settings; this Skill defines the decision criteria and evidence requirements that prevent those failures.

Core Features & Use Cases

• Approval mode definitions: Clarifies interactive, restricted-autonomous, and sandbox-autonomous scopes and the operations that must always remain human-gated.
• Precondition checks and artifacts: Specifies five required preconditions (sandboxed mode, implementation phase, documented execution profile, credentials out of Git, configured audit path) and the related artifacts and backlog stories needed to enable autonomy.
• Audit and maintenance guidance: Describes required audit evidence location, content, and story-based governance to ensure every autonomous run is reviewable and reversible.
• Use case: A team moving to CI-like sandboxed runs can use this Skill to create the execution-profile, record an approved HUS-* decision, and configure runtime audit recording before enabling restricted-autonomous tasks.

Quick Start

Verify the five autonomous preconditions, create and approve a HUS-* story recording the desired approval_mode and execution_mode, and then write the execution-plan/runtime/execution-profile.yaml to enable the approved autonomous boundaries.