self-review
Pass
Audited by Gen Agent Trust Hub on Aug 4, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted user-supplied manuscripts which creates an indirect prompt injection surface, potentially allowing embedded text to influence agent behavior through its available tools. \n
- Ingestion points: The agent reads multiple files from a user-controlled
review_packet/directory using standard file tools. \n - Boundary markers: The instructions provide clear logical boundaries by defining a strict 'Clean-Room' report structure and forbidding the use of non-manifested data or prior chat memory as evidence. \n
- Capability inventory: The skill has access to
Bash,Write,Edit,Read,Glob, andGreptools, and executes its own Python validation script via shell. \n - Sanitization: The skill does not perform automated sanitization of the manuscript content or embedded instructions. \n- [SAFE]: The skill contains a security-oriented validation script (
check_self_review_packet.py) that verifies the review manifest for accidental credential exposure and absolute path references to prevent directory traversal. \n- [SAFE]: The configuration instructions explicitly follow best practices by mandating the use of environment variables for API key management and prohibiting the storage of secret values in plaintext configuration files.
Audit Metadata