The Agent Skills Directory

[SAFE]: The skill includes a 'Snapshot And Protect' workflow that requires the agent to confirm the skill path and preserve existing user changes using version control or timestamped backups before performing any edits.
[SAFE]: A static audit step is mandated to check for high-risk patterns, specifically capping the audit score if hardcoded secrets or destructive actions without approval boundaries are detected.
[COMMAND_EXECUTION]: The skill facilitates the evaluation of skills within various harnesses (like Codex or Claude Code). It mitigates execution risks by categorizing tasks by risk level and requiring human confirmation for any action that changes safety gates, permissions, or external communications.
[EXTERNAL_DOWNLOADS]: The documentation references several external GitHub repositories from well-known technology organizations and the vendor's own organization as conceptual anchors. These references are used to inform the optimization logic and do not involve the automated download or execution of untrusted scripts.
[PROMPT_INJECTION]: As the skill is designed to process untrusted data (existing skill documents), it addresses potential indirect prompt injection (Category 8) by defining clear ingestion points, maintaining a capability inventory limited by human gates, and applying sanitization through its static audit rubric.

skillplus-improve