perf
Pass
Audited by Gen Agent Trust Hub on Mar 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for building and testing the project (e.g., 'Run the project's test suite', 'Run build/compile') during its 'Self-Healing Validation' phase. It also directs the agent to access and write to paths within '~/.claude/projects/' to record telemetry data.
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted codebase data and performs sensitive actions based on that data. 1. Ingestion points: Reads the entire codebase including manifest files and source code files (Phases 1-7). 2. Boundary markers: None are specified to isolate the ingested code content or instruct the agent to ignore embedded instructions. 3. Capability inventory: Modification of project source code (Self-Healing phase) and execution of system commands (Phase 9). 4. Sanitization: No sanitization or validation of the analyzed codebase content is performed.
Audit Metadata