retro
Pass
Audited by Gen Agent Trust Hub on Jul 28, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions describe the use of local command-line tools, such as
eslint --print-config, to inspect project-specific configurations. This is a legitimate development workflow used to avoid duplicate or conflicting rule generation. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted conversation history to stage potential project improvements, creating a potential surface for indirect prompt injection.
- Ingestion points: Conversation history (Step 1) and
.ai/learnings/files (Step 2). - Boundary markers: Explicit user consent requirement (Step 4) and instructions to use only externally observable evidence.
- Capability inventory: Native file system writes, project-local rule generation, and agent native memory updates.
- Sanitization: All modifications are gated by a manual user approval step.
Audit Metadata