The Agent Skills Directory

[SAFE]: The skill is a defensive utility that uses a PreToolUse hook to validate bash commands against a set of safety rules defined in rules.json. It provides different operating modes (Careful, Freeze, Unfreeze) to manage operational risk.
[COMMAND_EXECUTION]: The skill includes a local bash script (bin/check-dangerous.sh) used as a security gatekeeper. It analyzes commands using standard utilities like jq and grep without executing the commands themselves. The script provides feedback to the agent to prevent execution of blocked patterns.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by interpolating user-controlled command strings into its warning messages in SKILL.md. While this is inherent to the tool's purpose of reporting what it blocked, it creates a potential channel for malicious commands to attempt to influence the agent's subsequent logic.
Ingestion points: Command strings captured by the PreToolUse hook in SKILL.md and bin/check-dangerous.sh.
Boundary markers: None present in the SKILL.md warning template.
Capability inventory: The skill has the capability to analyze, block, and warn about shell commands using its internal logic and configuration files.
Sanitization: Validation is performed through regex pattern matching in rules.json to identify and stop dangerous commands before they are processed by the warning template.

guard