The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, specifically GitHub Pull Requests and Issues, which could potentially contain malicious instructions.\n
Ingestion points: Data is gathered via gh pr list and gh issue list commands in SKILL.md and the workflow script.\n
Capability inventory: The skill possesses significant capabilities including the execution of shell commands (git, gh, grep), script execution (bash, python3, uv), and the Workflow tool for parallel processing.\n
Boundary markers: The workflow script uses structured prompts to guide sub-agents, but doesn't explicitly wrap external data in distinct delimiters to separate it from instructions.\n
Sanitization: The skill includes a mandatory counter-review step in SKILL.md that requires the agent to manually verify all high-severity findings using direct commands, providing a human-in-the-loop style validation layer against malicious input.\n- [COMMAND_EXECUTION]: The skill utilizes common development and repository management tools (gh, jq, find, grep, git, wc, tr) to perform its audit functions. These operations are transparent and consistent with the skill's stated purpose.\n- [REMOTE_CODE_EXECUTION]: The skill executes several internal scripts (check_marketplace.sh, check_doc_skill_lists.py, security_scan.py) to validate the repository's state. These scripts are assumed to be part of the skill's own distribution or the repository being audited.\n- [DATA_EXFILTRATION]: The skill is specifically designed to locate and report sensitive information such as PII and secrets. It includes explicit instructions on 'Anti-target' behavior to prevent these discovered secrets from being inadvertently published in repository configuration files like .gitleaks.toml.

marketplace-health-check