fable-like-implementation
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill explicitly directs the agent to execute build, lint, and test commands found within the target repository to establish a 'Verification Baseline'.
- Evidence: Found in
SKILL.mdunder 'Operating posture' ('run the build/lint/test commands to learn whether the baseline passes today') and 'Stage 2' ('The exact commands that build/lint/test this repo and whether each passes right now (run them)'). Execution of arbitrary local scripts poses a risk if the repository contains malicious configuration. - [PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted input from both the user and the repository codebase to generate its plans.
- Ingestion points: The workflow involves reading the user's initial request (
Stage 1) and performing wide scans of codebase files, manifests, and documentation (Stage 2). - Boundary markers: The skill includes a hard rule stating that implementation should not begin until a plan is complete and the user explicitly approves it, providing a human-in-the-loop checkpoint.
- Capability inventory: The agent is authorized to execute shell commands for verification and utilize sub-agents for network-based documentation research.
- Sanitization: There are no explicit instructions for sanitizing or escaping content retrieved from the codebase before it is incorporated into the LLM's reasoning context.
Audit Metadata