neo-harness-project
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted content from the target project's codebase to generate documentation and execute validation scripts.
- Ingestion points: The agent reads project root directories, configuration files like package.json, and existing documentation as specified in SKILL.md and references/project-assessment.md.
- Boundary markers: The skill enforces a strict Write Preview step in SKILL.md (Step 3), requiring the agent to present all planned actions and justifications for user approval before execution.
- Capability inventory: The skill writes new files to the project directory and executes commands derived from project settings, as detailed in references/generation-validation.md.
- Sanitization: The skill instructions focus on evidence-based generation and template cleanup but do not detail specific filtering for adversarial instructions within ingested data.
- [COMMAND_EXECUTION]: The skill is designed to identify and execute existing project commands for verification.
- Evidence: SKILL.md (Step 5) and assets/templates/AGENTS.md.template instruct the agent to run verified commands and unified verification scripts. The skill explicitly forbids adding unrequested third-party dependencies or external connections.
Audit Metadata