The Agent Skills Directory

[SAFE]: The skill implements a robust software delivery workflow with multiple safety checks.
[SAFE]: Uses a cryptographic fingerprinting mechanism in scripts/plan-header.mjs to verify that the implementation plan matches the authorized issue requirements and hasn't been tampered with or drifted from the intended base commit.
[SAFE]: Enforces a strict delivery gate through references/delivery-gate.md that requires independent review from a different model provider (e.g., a different lab) to prevent self-review bias and ensure quality.
[SAFE]: Explicitly limits the agent's authority to the authorized scope of the assigned issue and instructs the agent to stop and escalate to a human/planner if any engineering decisions or plan gaps are encountered.
[SAFE]: The included Node.js scripts (delivery-status.mjs, plan-header.mjs) are diagnostic and validation tools that process local data without performing network operations, accessing sensitive host credentials, or executing arbitrary input.

execute