planning-with-files
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves performing research via web searches and storing findings in a
notes.mdfile (as seen in Example 1 and Example 3). This data is later read back into the agent's context for synthesis and delivery. Maliciously crafted content from external websites could attempt to hijack the agent's logic during the retrieval or synthesis phases.\n - Ingestion points: The
notes.mdfile serves as a storage point for untrusted external data.\n - Boundary markers: No specific delimiters or instructions to ignore embedded commands are recommended in the templates or core workflow.\n
- Capability inventory: The skill assumes capabilities for file read/write operations and web searching.\n
- Sanitization: There are no instructions provided for sanitizing or validating external input before it is processed by the agent.\n- [COMMAND_EXECUTION]: The skill instructs the agent to use standard file system operations (read, write, edit, and append) to manage its 'working memory' on disk. These actions are aligned with the skill's stated purpose of improving long-term task coherence.
Audit Metadata