writing-docs
Warn
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: MEDIUMOBFUSCATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [OBFUSCATION]: The reference files
reference/mdx-patterns.md,reference/book-style.md, andreference/resources-style.mdcontain Zero-Width Space (U+200B) characters immediately preceding code block markers (e.g.,```ts). While these are used for formatting purposes to demonstrate MDX syntax without terminating the enclosing code block, they are invisible characters that modify the file content. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process code diffs from external sources (such as pull requests) to update documentation. This creates a data ingestion point for untrusted content. A malicious code change or comment could include instructions intended to override the agent's behavior during documentation generation. The skill currently lacks explicit boundary markers or sanitization instructions for this ingested data.
- [COMMAND_EXECUTION]: The skill instructions in
SKILL.mdidentify local development scriptsyarn prepandyarn lint:content. Although it explicitly forbids the agent from running them manually, their presence in the instructions defines them as part of the skill's operational context.
Audit Metadata