scene-splitter
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied stories, which represents an attack surface for indirect instructions. This is assessed as safe due to strong human-in-the-loop controls.\n
- Ingestion points: User-provided plain English stories processed in Stage 1 and Stage 2 of the workflow.\n
- Boundary markers: The skill employs a structured JSON output and a reconstruction check (Stage 3.5), though it lacks explicit 'ignore instructions' delimiters within the story text itself.\n
- Capability inventory: The skill possesses the capability to write JSON files to the local 'stories/{slug}/' directory (Stage 5).\n
- Sanitization: A 'HARD GATE' (Stage 4) requires manual user review and approval of the generated scene plan before any file-write operations occur.\n- [SAFE]: The skill implements defensive programming by sanitizing the story 'slug' with a strict regex (^[a-z0-9-]+$) to prevent path traversal when writing output files.\n- [SAFE]: No indicators of prompt injection, obfuscation, or unauthorized remote code execution were found.
Audit Metadata