The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides a GitHub Actions template that uses the GitHub API to fetch a script from a private repository and executes it via bash. While the source is a well-known service, the execution of remote content is a sensitive operation.
[COMMAND_EXECUTION]: The skill includes bash scripts that execute local tools such as grep, ripgrep, and jq to perform deep scanning of the file system and artifact trees.
[EXTERNAL_DOWNLOADS]: The CI workflow configuration fetches content from external GitHub repositories. As these references target a well-known service (GitHub), this is documented as a standard operational pattern.
[DYNAMIC_EXECUTION]: The skill's CI template decodes a Base64-encoded payload from a remote API call and writes it to a file (gate.sh) for immediate execution.
[PRIVILEGE_ESCALATION]: The GitHub Actions configuration template includes the use of sudo to install system dependencies (ripgrep and jq) in the CI environment.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad ingestion surface, as it is designed to scan arbitrary files including HTML, JSON, and source code, which could contain malicious instructions.
Ingestion points: The TARGET tree defined in the scanning scripts in SKILL.md.
Boundary markers: Absent; the skill performs raw analysis of the provided data.
Capability inventory: Uses bash, grep, ripgrep, and jq for file processing and verification.
Sanitization: Absent; the logic focuses on leak detection rather than input sanitization for the agent context.

enforce-redaction-gate