write-like-me
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests a corpus of external data, such as blog posts or repository files, to extract stylistic patterns. This untrusted data source could contain hidden instructions (e.g., within markdown comments or metadata) designed to influence the agent's behavior during analysis or generation.\n
- Ingestion points: The
init,rewrite, andeditmodes process files from user-defined paths, repository locations, or pasted text provided in the conversation context.\n - Boundary markers: The instructions do not prescribe the use of mandatory boundary markers (such as XML tags or unique delimiters) to isolate the corpus content from the agent's instructions.\n
- Capability inventory: The agent has access to read and write files within specific user-managed directories (
~/.write-like-me/) and can executegitcommands for provenance verification.\n - Sanitization: The instructions focus on linguistic analysis and do not include steps to sanitize or escape potential instructions embedded in the input text.\n- [COMMAND_EXECUTION]: The skill directs the agent to run shell commands like
git log --diff-filter=Ato verify the history and authorship of files within the provided corpus. While used for legitimate provenance triage, it involves executing logic based on the state of an external repository.
Audit Metadata