reproducibility-audit
Pass
Audited by Gen Agent Trust Hub on Apr 25, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from the user's repository for analysis.
- Ingestion points: The agent is instructed to read repository files and structure to audit environment management, data versioning, and documentation (SKILL.md Stage 1 and 2).
- Boundary markers: The instructions lack explicit delimiters or warnings to ignore embedded instructions within the files being audited.
- Capability inventory: The agent has the ability to read files from the filesystem and write the final audit report to the ~/phd-log/ directory (SKILL.md Stage 5).
- Sanitization: No sanitization or validation of the repository content is performed before processing.
Audit Metadata