The Agent Skills Directory

[COMMAND_EXECUTION]: The skill describes a wrapper for managing CLI operations with risk-based filtering. It uses pattern matching to authorize or deny shell commands based on user-defined risk thresholds.
[EXTERNAL_DOWNLOADS]: Recommends the installation of standard utilities yq and jq via Homebrew, a well-known and trusted package manager.
[PROMPT_INJECTION]: No direct prompt injection or bypass attempts were found. The skill includes an indirect prompt injection surface by design, as it parses security policies from external SAFEGUARDS.md files located in plugin directories. A malicious plugin could theoretically provide a deceptive policy to misclassify dangerous commands as safe.
Ingestion points: SAFEGUARDS.md files in various plugin subdirectories.
Boundary markers: Uses YAML blocks within HTML comments in markdown files.
Capability inventory: Controls the allow[] and deny[] lists for CLI command execution.
Sanitization: Relies on yq and jq for parsing structured YAML data.
[SAFE]: The skill acts as a security enhancement for CLI operations and does not contain malicious code or instructions.

safe-mode