The Agent Skills Directory

[SAFE]: The skill operates entirely locally using static analysis techniques to evaluate other skills. It reads markdown files and uses simple Python scripts for counting steps and calculating heuristic reliability scores.
[SAFE]: All file access is scoped to the skill being analyzed (via a skill directory path) and the agent's own internal skill directories (~/.claude/skills/, etc.). No attempts to access sensitive system directories, credentials, or environment variables were detected.
[SAFE]: The Python scripts inventory.py and trigger_lint.py are dependency-free and perform deterministic text processing (regex matching and simple arithmetic). They do not spawn subshells, perform network requests, or use dangerous functions like eval() or exec().
[SAFE]: The workflow involves a human-in-the-loop for any modifications. Proposed fixes (such as description rewrites) are only applied to the target skill's SKILL.md after explicit user confirmation.
[SAFE]: No obfuscation, hidden instructions, or persistence mechanisms were found. The skill's behavior is consistent with its stated purpose of improving agent reliability through auditing.

skill-reliability