anti-defensive-writing
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides a series of bash search commands in the 'Detection pass' section of 'SKILL.md'. These commands use 'grep' to identify specific writing patterns within LaTeX source files. This represents local file access and execution for the skill's core analytical functionality.
- [EXTERNAL_DOWNLOADS]: The 'UPSTREAM.md' file references the skill's source on GitHub and provides a 'curl' command for documentation and manual refresh purposes. The URL targets a well-known service for source code hosting.
- [INDIRECT_PROMPT_INJECTION]: As the skill is designed to ingest and transform user-supplied scientific manuscripts, it possesses an attack surface for indirect prompt injection. * Ingestion points: The agent reads and processes 'manuscript.tex', Supplementary Information, legends, and table notes ('SKILL.md'). * Boundary markers: The skill suggests ignoring LaTeX comments but lacks explicit structural delimiters to separate untrusted manuscript content from the agent's instructions. * Capability inventory: The skill is capable of file reading, text modification, and executing 'grep' commands on the local file system. * Sanitization: The skill's analysis is constrained to specific textual regex patterns, which inherently limits the impact of arbitrary content within the manuscript.
Audit Metadata