paper-section-writer
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external markdown and JSON artifacts as its primary inputs. It effectively mitigates risk by enforcing mandatory 'sentinels' ([MODELER INPUT NEEDED]) for high-stakes judgments like physical domain meanings, innovation claims, and method selection rationales. This design ensures the AI cannot fabricate critical content and maintains clear boundaries between AI-drafted mechanics and human-authored judgment.
- [DATA_EXFILTRATION]: The skill does not contain any network operations (e.g., curl, wget) or attempts to access sensitive system files like credentials, environment variables, or SSH keys. All operations are confined to local project directories.
- [COMMAND_EXECUTION]: There are no shell command executions, subprocess spawns, or system-level modifications present in the instructions.
- [OBFUSCATION]: The content is presented in clear markdown and YAML without the use of Base64 encoding, zero-width characters, homoglyphs, or other techniques designed to hide malicious intent.
- [DYNAMIC_CONTEXT_INJECTION]: The skill does not use the dynamic context injection syntax (
!command) and operates purely as a set of instructions for text generation and structured reporting.
Audit Metadata