manuscript-import
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external, untrusted content (manuscript drafts) which could contain embedded instructions to influence the agent's behavior.
- Ingestion points: The skill reads the 'whole draft' of an existing manuscript to inventory structure, characters, and plot (documented in SKILL.md under 'Inventory' and 'Principle').
- Boundary markers: The instructions do not specify explicit delimiters (like XML tags or specific markers) to isolate the prose content from the agent's core instructions, increasing the risk of the agent misinterpreting manuscript text as command overrides.
- Capability inventory: The skill has the capability to write to multiple project files, including
import/IMPORT_PLAN.md,import/EXTRACTION_LOG.md, andglossary/GLOSSARY.md, and invokes several other management tools (context-manager,canon-manager,timeline, etc.). - Sanitization: No specific sanitization or filtering of the ingested prose is mentioned before it is processed or used to populate project files.
Audit Metadata