The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill incorporates a dynamic lookup mechanism in LOOKUP.md to retrieve information from external libraries.\n
Ingestion points: Untrusted data enters the agent's context through the query-docs tool, which fetches third-party documentation content based on user-provided library names and queries.\n
Boundary markers: The instructions lack specific delimiters or system-level directives to prevent the agent from following instructions that might be embedded within retrieved documentation strings.\n
Capability inventory: The skill has access to the Write and Bash tools (as defined in DOCS-WRITE.md), which could be misused if the agent is successfully manipulated by malicious content in the fetched documentation.\n
Sanitization: There is no explicit process described for sanitizing or filtering the content of external documentation before it is processed by the agent.\n- [COMMAND_EXECUTION]: The DOCS-WRITE.md file instructs the agent to execute shell commands to perform maintenance tasks.\n
Evidence: The agent is specifically directed to run yarn prettier --write <file-path> to format files after editing. This behavior relies on the Bash tool and involves the execution of external binaries on the local filesystem.

docs-pro