project-improver
Pass
Audited by Gen Agent Trust Hub on Apr 12, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The file references/evals.yaml contains a test case specifically designed to ensure the model rejects malicious instructions, such as shell command injection. This is a safety feature for benchmarking.\n- [DATA_EXFILTRATION]: The skill instructions in SKILL.md require the agent to redact secrets, tokens, and credentials from all generated summaries and plans, mitigating the risk of credential exposure.\n- [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection because it reads and processes untrusted data from external repositories.\n
- Ingestion points: Analysis of current codebase, docs, and project plans in SKILL.md (Phase 0).\n
- Boundary markers: Grounding rules are established, but there are no explicit delimiters or instructions to ignore embedded commands within the analyzed data.\n
- Capability inventory: The skill can implement code changes and modify the repository structure in Phase 6.\n
- Sanitization: No explicit sanitization or filtering of instructions found within processed files is mentioned.\n- [COMMAND_EXECUTION]: The skill invokes a local script utilities/skill-builder/scripts/record_skill_feedback.py to log user feedback at the end of the workflow.
Audit Metadata