vision
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script (
python3 docs/ux/lint.py). This command is used to validate UX changes and check for architectural drift, but it involves executing code within the user's project environment. - [PROMPT_INJECTION]: The skill programmatically modifies project-level instruction files (such as
CLAUDE.md,AGENTS.md, orGEMINI.md) to install a "vision-alignment rule". This rule forces the agent to check all future features against thevision.mddocument, effectively injecting persistent behavioral constraints into the agent's core instructions. - [PROMPT_INJECTION]: The skill ingests data from multiple project files including
README.md,ARCHITECTURE.md, and source code to derive its vision document. It lacks explicit sanitization or boundary markers for this ingested data, creating a surface for indirect prompt injection if the project files contain malicious instructions. - Ingestion points:
README.md,ARCHITECTURE.md,ROADMAP.md,CHANGELOG.md,docs/*, source code entry points,package.json,pyproject.toml,Cargo.toml. - Boundary markers: Absent.
- Capability inventory: File writing (
docs/ux/vision.md,README.md), modification of agent instructions (CLAUDE.md), and shell command execution (python3 docs/ux/lint.py). - Sanitization: Absent.
Audit Metadata