The Agent Skills Directory

[SAFE]: The skill implements a 'Correctness Doctrine' across all workflows, instructing agents to prioritize source code and real-time command output as the authoritative source of truth, effectively mitigating risks of acting on potentially stale or malicious documentation.
[SAFE]: All automated workflows use standard developer tools like git, GitHub CLI (gh), and SSH for their intended project management and auditing purposes, with operations restricted to the user's project context and governed by explicit user-supplied parameters.
[SAFE]: The multi-layered 'Review Ladder' (L1-L3) ensures that code implementation and planning decisions are reviewed by independent model instances and adversarial perspectives before completion, providing strong validation of automated tasks.
[SAFE]: Deployment logic in the install script and manifest generator uses standard shell operations and hash-based verification to manage files securely, preventing accidental overwrites or unauthorized modifications to the workflow environment.

using-workflows