The Agent Skills Directory

[METADATA_POISONING]: The skill-report.json metadata file contains a pre-emptive security audit and risk level assertion ("risk_level": "safe") which attempts to influence the safety evaluation of the skill.
[DATA_EXFILTRATION]: The discover_env.py script identifies and logs the paths of environment configuration files (e.g., .env, .env.*) to assess exposure risks, which is consistent with the skill's primary purpose.
[DATA_EXFILTRATION]: The fetch_vulns.py script transmits discovered package names and versions to the api.osv.dev service, a well-known vulnerability database, to retrieve security advisory data.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project manifest files and interpolates their content into reports that suggest shell commands and Python code for remediation.
Ingestion points: scripts/discover_env.py parses package.json, requirements.txt, Gemfile, and go.mod files from the local repository.
Boundary markers: The skill does not use specific delimiters or instructions to treat ingested package data as untrusted when generating remediation snippets.
Capability inventory: The skill uses Bash, Write, and WebFetch tools. The generated reports suggest executing package manager commands (e.g., npm install, pip install) and running Python test scripts that use subprocess.run.
Sanitization: Extracted data such as package names and version strings are used directly in report templates without validation or escaping, creating a surface for injection into the generated report content.

security-analyzer