cn-thesis-references
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill employs a manifest-driven architecture and a specific routing protocol that defines the agent's operational boundaries, effectively limiting the risk of behavioral drift.
- [SAFE]: The instructions contain explicit guardrails against the fabrication of academic metadata (e.g., DOIs, URLs, authors), which promotes data integrity and prevents hallucination.
- [SAFE]: All document modifications are performed via local scripts with a 'dry-run' default, ensuring user data is protected and not overwritten without explicit flags.
- [PROMPT_INJECTION]: The skill processes untrusted user-supplied data in the form of documents and bibliography files, presenting a surface for indirect prompt injection.
- Ingestion points: User-provided thesis documents (.docx) and reference metadata files (.bib, .ris, JSON) as specified in README.md.
- Boundary markers: The Routing Protocol and manifest-driven architecture defined in SKILL.md provide structural delimiters for agent processing.
- Capability inventory: The skill utilizes file reading and writing capabilities through the scripts/apply_docx_references.py script as described in the workflow.
- Sanitization: The workflow focuses on bibliography format validation and the use of manual placeholder markers ([TBD: ...]) for missing data rather than autonomous completion.
Audit Metadata