The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a workflow that ingests external implementation plans to drive subagent actions, which creates a surface for indirect prompt injection. Ingestion points: Plan files read in Step 1 of SKILL.md. Boundary markers: None explicitly defined in subagent prompts to isolate task instructions. Capability inventory: The system executes shell commands for testing/building, performs Git commits, and dispatches new subagents. Sanitization: Plan content is used directly without escaping. This design is intrinsic to the skill's purpose as an automation tool and is balanced by the required manual and automated review steps.
[COMMAND_EXECUTION]: The skill references standard development commands for verification. Evidence: Use of pytest, pnpm test, and pnpm build in the execution checklist.

executing-plans