The Agent Skills Directory

[SAFE]: No security issues detected. The skill's operations are limited to discovering and editing instruction files, which matches its stated purpose. The skill is adapted from a trusted repository (Anthropics) and contains no code for unauthorized network access or credential theft.\n- [COMMAND_EXECUTION]: The skill uses a standard shell command (find) to locate instruction files within the repository. This is a benign discovery operation with limited scope and appropriate output truncation.\n- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it processes untrusted instruction files (Ingestion points: AGENTS.md, CLAUDE.md, .agents/instructions.md, ~/.config/opencode/AGENTS.md). While boundary markers and sanitization for these inputs are absent, the skill's capability inventory is limited to project discovery and guided file updates which require user confirmation. This surface is inherent to the skill's primary purpose and is considered safe given the workflow controls.

agents-md-improver