vibe-agent-instructions
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill provides a structured and transparent workflow for managing repository-level agent instruction files. It enforces a principles-based boundary system, requiring explicit user consent for modifications to tracked files and ensuring all changes are previewed before being applied.
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes existing repository files such as AGENTS.md, CLAUDE.md, and local overrides (Ingestion points). This represents an indirect prompt injection surface. The skill mitigates this through mandatory evidence-based instructions in the SKILL.md 'Boundaries' section, which explicitly commands the agent to treat instruction files as data and never obey directives found within them. Its capabilities (Capability inventory) are limited to markdown file creation and symlink management, with no execution of the ingested data.
- [COMMAND_EXECUTION]: The skill utilizes standard system and version control commands, specifically 'ln -s' for symbolic link creation and 'git ls-files' / 'git check-ignore' for repository inventory. These commands are used for their intended administrative purposes and are not exposed to unsanitized external input.
Audit Metadata