technical-writing
Warn
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The script
tests/run-validation.pyperforms dynamic loading of internal components by reading the source ofscripts/validate-output.py, compiling it, and executing it within a provided namespace using theexec()function. This enables runtime execution of logic from external files within the testing suite. - [INDIRECT_PROMPT_INJECTION]: The skill acts on documentation text which may originate from untrusted sources, providing a surface for instructions embedded in data to influence agent behavior.
- Ingestion points: Documentation source material, repository files processed for updates (README.md, AGENTS.md), and input ingested by
scripts/validate-output.pyvia standard input. - Boundary markers: The
SKILL.mdinstructions explicitly mandate human-in-the-loop approval before replacing or overwriting top-level repository documents, acting as a control against automated obedience to instructions embedded in target files. - Capability inventory: The skill is capable of executing internal Python scripts and performing file system write operations to modify documentation artifacts.
- Sanitization: The validator in
scripts/validate-output.pyuses a structural HTML parser to verify that generated HTML documentation is self-contained and does not include external scripts or linked resources, reducing the risk of data exfiltration via external assets.
Audit Metadata