evolve
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from the
_workspace/directory and user-provided feedback to automatically modify its own instructions and those of other agents in the.claude/directory. - Ingestion points: Data from the
_workspace/directory (recent execution artifacts),CLAUDE.md, and direct user feedback. - Boundary markers: No explicit markers are used to isolate or delimit untrusted data during the feedback generalization process.
- Capability inventory: The skill has the capability to write and modify markdown files within the
.claude/agents/and.claude/skills/directories, which directly define agent behaviors and constraints. - Sanitization: No programmatic sanitization or validation of the ingested data is performed before it is generalized and interpolated into agent instructions.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute local shell commands to retrieve repository history for analysis.
- Evidence: The skill uses
git log --oneline -- .claude/ CLAUDE.mdto identify changes in the configuration files since the initial setup.
Audit Metadata