self-healing
Pass
Audited by Gen Agent Trust Hub on Mar 27, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill facilitates persistent indirect prompt injection by prioritizing user-provided corrections ('User corrections override everything') and storing them in long-term memory. This allows malicious instructions to be embedded in memory files (e.g., MEMORY.md, topic files) and automatically re-loaded into the agent's context in future sessions. Ingestion points include user corrections and analyzed codebase patterns; no boundary markers are specified to delimit this untrusted data; the agent has the capability to write files and create executable scripts; and no sanitization process is described to prevent instruction injection within the stored data.
- [COMMAND_EXECUTION]: The instruction set in 'skill-creation-guide.md' encourages the agent to create new functional skills which can include executable shell scripts (e.g., 'validate.sh'). This capability allows the agent to generate and execute arbitrary system commands based on patterns it identifies during the session.
Audit Metadata