docwriter-style-generator
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes bundled Node.js scripts (
scripts/save-source.mjsandscripts/analyze-style.mjs) to automate the retrieval, sanitization, and statistical analysis of writing samples provided by the user. - [DYNAMIC_EXECUTION]: The skill programmatically generates a complete AI agent skill (including
SKILL.mdand reference JSON files) at runtime. These files are written to the agent's skill loading directories (~/.claude/skills/and~/.agents/skills/), which effectively modifies the agent's future behavior based on the learned style profile. - [EXTERNAL_DOWNLOADS]: The skill provides an option to fetch writing samples from remote URLs. The
scripts/save-source.mjsscript uses thefetchAPI to download content, which is then stripped of HTML chrome and stored locally for analysis. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external URLs and local files. While this creates a potential surface for indirect prompt injection, the risk is mitigated by a multi-agent filtering pass, a human-in-the-loop calibration step where the user must approve generated propositions, and explicit instructions in the generated output for the agent to prioritize user rules over the learned style examples.
Audit Metadata