dependency-auditor
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [METADATA_POISONING]: The skill documentation and source code headers across multiple files (SKILL.md, dep_scanner.py, license_checker.py, upgrade_planner.py) consistently claim the author is the 'Claude Skills Engineering Team' or 'Claude Skills Team'. This is deceptive as the skill is provided by an external third-party author ('alirezarezvani'). This impersonation of official platform teams can lead to misplaced trust in the skill's safety and capabilities.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external manifest and lockfiles (e.g., package.json, requirements.txt, go.mod) to generate reports that are then analyzed by the AI agent. If these files are controlled by an attacker, they can include malicious instructions in metadata fields like package names or descriptions that may trigger the agent.
- Ingestion points: Multiple parsing functions in
dep_scanner.py(e.g.,_parse_package_json,_parse_requirements_txt), and_scan_project_dependenciesinlicense_checker.py. - Boundary markers: The reports use text-based visual delimiters (e.g., '=====') but do not include explicit 'ignore instructions' directives for the AI agent regarding the parsed dependency data.
- Capability inventory: The skill allows writing reports to the local file system using the
--outputargument. - Sanitization: There is no evidence of string sanitization or escaping of dependency metadata before it is interpolated into the text and JSON reports used by the agent.
Audit Metadata