auditing-npm-dependencies
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes vulnerability advisories from the external npm registry. Maliciously crafted package metadata or advisory descriptions could theoretically contain prompt instructions designed to influence the agent's behavior during the audit process.
- Ingestion points: The
scripts/audit_npm.pyscript parses JSON output from thenpm auditcommand. - Boundary markers: Absent. The skill does not implement specific delimiters or 'ignore' instructions for the data retrieved from the registry.
- Capability inventory: The skill is authorized to use
Bashfornpmandpython3commands and can read project files. - Sanitization: Absent. The script maps external data directly to report fields without sanitizing for potential LLM instructions.
- [COMMAND_EXECUTION]: The skill utilizes
subprocess.runwithinscripts/audit_npm.pyto execute thenpm auditcommand. While this is the intended and necessary behavior for the skill's purpose, it represents an automated interaction with the system shell.
Audit Metadata