The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from external primary sources such as official documentation, source code repositories, and API specifications. This introduces a risk where malicious instructions embedded in those sources could influence the behavior of the subagents.
Ingestion points: Step 2 (Survey directions) and Step 4 (Dispatch direction subagents) retrieve data from external URLs and primary sources.
Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded commands within the gathered research material.
Capability inventory: The skill has the ability to write multiple files to the local file system (~/research/), spawn subagents using runs.run, and execute generated scripts via workflowScriptPath.
Sanitization: There is no evidence of sanitization or filtering applied to the content retrieved from primary sources before it is processed by subagents.
[DYNAMIC_EXECUTION]: The pipeline instructions require the agent to generate and execute local scripts to manage complex subagent workflows.
Evidence: Step 4 explicitly instructs the agent to "write the workflow to a script file and pass workflowScriptPath" to avoid pitfalls with nested template literals. This involves the runtime creation and execution of executable content.

research