The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions as a conductor for sub-agents, creating an attack surface where it ingests untrusted data in the form of status reports, PR links, and evidence from those agents via the createRun tool (SKILL.md). The skill possesses significant capabilities including spawning new agents (createAgent), executing shell commands (execute), and performing GitHub operations (gh). To mitigate the risk of being misled by sub-agent output, the instructions mandate a 'Final QA' phase and direct verification of all claims against the repository's main branch before the conductor considers a task finished.
[DYNAMIC_EXECUTION]: The skill utilizes the workflows.create tool (SKILL.md) to schedule deferred tasks. This involves generating a string of JavaScript code that is executed at a later time to trigger reporting wake-ups. While these are based on static templates for coordination purposes, they represent a dynamic code generation and execution pattern.

conduct