self-improving-agent
Pass
Audited by Gen Agent Trust Hub on Jul 18, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill implements an architecture that creates a surface for indirect prompt injection by processing and promoting data from
MEMORY.md, which contains untrusted input. - Ingestion points:
MEMORY.mdand topic files are ingested by thememory-analystagent for pattern recognition. - Boundary markers: The skill does not specify the use of boundary markers or delimiters to isolate untrusted content during the analysis or promotion phase.
- Capability inventory: The skill has the capability to write to and modify project rule files such as
CLAUDE.mdand.claude/rules/, which directly influence agent behavior. - Sanitization: There is no evidence of sanitization, filtering, or validation of the captured memory content before it is promoted to persistent project-level instructions.
- [COMMAND_EXECUTION]: The installation section provides instructions for executing a local script (
./scripts/codex-install.sh) which may perform arbitrary commands on the user's system.
Audit Metadata