fable-loop
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by integrating untrusted external data into the agent's decision-making and execution pipeline.
- Ingestion points: Stage 1, Step 2 instructions direct the spawning of 'research agents' to fetch documentation and perform web searches (SKILL.md).
- Boundary markers: The workflow includes multiple internal checkpoints such as an 'intent gate' before behavior changes (Stage 2, Step 2), a 'recall gate' for external facts (Stage 2, Step 5), and an 'authorization gate' for any outward-facing actions (Stage 2, Step 6).
- Capability inventory: The agent retains capabilities for surgical file edits, checklist execution, and running local verification tools like tests and linters in the main thread (Stage 2 and Stage 3).
- Sanitization: The workflow incorporates an 'Adversarial Verification' stage (Stage 3) where parallel subagents are prompted to refute the work and identify contradictions, acting as a defense-in-depth mechanism against potentially malicious or malformed input data.
Audit Metadata