The Agent Skills Directory

[COMMAND_EXECUTION]: The audit_marketplace.py script invokes git remote via subprocess.check_output to verify if the marketplace name aligns with the repository name. The command uses hardcoded arguments and does not incorporate unsanitized user input.\n- [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests and processes content from marketplace.json, plugin markdown files, and documentation.\n
Ingestion points: Reads marketplace.json, CLAUDE.md, and all registered agent, skill, and command markdown files.\n
Boundary markers: The audit report outputs findings directly to the agent's context without specific delimiters or warnings about embedded instructions.\n
Capability inventory: The script can execute git commands and has the capability to write to markdown files when the --fix flag is used.\n
Sanitization: Field extraction uses restrictive regex patterns, but the data is not sanitized against prompt injection techniques before being reported as a finding.

marketplace-audit