flow-prune-agent-setup
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a strictly managed administrative workflow for tidying agent configurations. It emphasizes user control, reversible edits, and format validation.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to read and process external configuration and instruction files, which serves as a potential surface for indirect prompt injection from data within the agent's environment.
- Ingestion points: Instruction files (e.g.,
AGENTS.md) and configuration/settings files (e.g.,settings.json) at various scopes (user, project, local). - Boundary markers: The skill does not explicitly define delimiters for untrusted content but instructs the agent to separate managed context from user context.
- Capability inventory: The agent is authorized to perform reversible edits to files and utilize owning installers or plugin managers to remove skills.
- Sanitization: The skill prescribes backing up settings, preserving unrelated keys, and validating resulting formats and effective behavior post-edit.
Audit Metadata