The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py executes the pip-audit command-line utility to identify known vulnerabilities. This is implemented using subprocess.run with a fixed list of arguments, which prevents command injection.
[EXTERNAL_DOWNLOADS]: The skill retrieves package metadata from the official PyPI JSON API at pypi.org. This is a necessary function for the skill's purpose and uses a well-known, trusted technology service.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface (Category 8) because it processes untrusted metadata from a public registry. \n
Ingestion points: Data is fetched from https://pypi.org/pypi/{package_name}/json in the query_pypi_metadata function within scripts/agent.py. \n
Boundary markers: The skill does not employ specific delimiters or instructions to the agent to disregard potential control sequences within the fetched metadata. \n
Capability inventory: The skill can execute local commands via subprocess.run and read local files via the open function. \n
Sanitization: The script parses API responses as JSON but does not sanitize or filter the content of metadata fields like descriptions or summaries before outputting them to the agent context.

performing-supply-chain-attack-simulation