The Agent Skills Directory

[SAFE]: The skill is primarily instructional, providing best practices for using the code.execute tool to optimize agent performance by reducing round-trips. It contains no executable malicious code, obfuscation, or persistence mechanisms.
[PROMPT_INJECTION]: The skill describes a workflow that ingests data from external tools (e.g., web.fetch) and uses it to drive subsequent logic and tool calls (e.g., web.search). While this is the intended purpose, it represents an indirect prompt injection surface where maliciously formatted external data could attempt to influence the logic within the code.execute block.
Ingestion points: Tools like web.fetch referenced in the example logic.
Boundary markers: None explicitly defined in the example code, though the skill notes that the host's hardline blocklist and approval gates still apply.
Capability inventory: Access to code.execute, which can further invoke tools like web.fetch and web.search via the global tools object.
Sanitization: The example uses JSON.parse(), but does not implement logic-level sanitization of the fetched content.

code-pipeline