self-evolution
Warn
Audited by Gen Agent Trust Hub on Aug 9, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill performs dynamic code generation and execution by patching existing scripts, creating new functions, and building new capabilities within 'Allowed Edit Directories' defined in the repo profile (Phase 1, 4). This self-modifying behavior allows the agent to persistently alter its own executable logic.
- [COMMAND_EXECUTION]: The agent is instructed to re-run shell commands or tests that previously failed to verify its fixes (Phase 5). This involves the use of the Bash tool to execute arbitrary code or scripts within the agent's environment.
- [PROMPT_INJECTION]: The skill exhibits an indirect injection surface by processing untrusted data (error messages, stack traces, and DOM snapshots) to generate code repairs (Phase 2).
- Ingestion points: Error messages, stack traces, and UI state snapshots (Phase 2).
- Boundary markers: Edits are restricted to directories defined in the 'self-evolution-profile.md' (Phase 0, 4).
- Capability inventory: Shell command execution via Bash and file modification via Write/Edit tools.
- Sanitization: No explicit validation or sanitization of external error content is performed before it is used to plan and execute code repairs.
Audit Metadata