The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches skill components and updates from the official prompt-security GitHub repository. The installation process includes rigorous security checks, including SHA256 checksum verification, artifact size limits to prevent Denial of Service, and zip-bomb/path-traversal protection.
[COMMAND_EXECUTION]: Utilizes system binaries like curl, jq, and gh to manage its lifecycle and perform reporting. Commands are executed with secure patterns, such as using single-quoted heredocs to prevent shell interpolation.
[DATA_EXFILTRATION]: Facilitates the submission of security incident reports to a centralized vendor-managed repository. This behavior is the primary intended function of the skill, requires explicit user consent for every submission, and follows documented privacy guidelines to avoid sharing sensitive data.
[PROMPT_INJECTION]: The skill processes external data regarding security incidents, presenting a potential surface for indirect prompt injection.
Ingestion points: Incident data gathered by the agent for report generation in SKILL.md and reporting.md.
Boundary markers: Documentation provides clear instructions on data sanitization and "DO NOT include" lists for sensitive information.
Capability inventory: The skill has the ability to write local files and perform network requests via the GitHub CLI (gh).
Sanitization: Implements secure shell scripting practices, specifically using literal heredocs to ensure untrusted content is not executed during file writing.

clawtributor