The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core workflow of processing external documentation.
Ingestion points: Subagents are instructed to read a design document from docs/plans/<feature>/design.md as the primary source for implementation planning in Phase 3.
Boundary markers: The prompt provided to subagents lacks explicit delimiters or "ignore embedded instructions" warnings for the content of the design document.
Capability inventory: The orchestrated subagents have access to powerful tools including file system modification, plan execution, and shell access for testing (npm test) and git operations.
Sanitization: No evidence of validation or sanitization of the design document's content is present before it is interpolated into subagent tasks.
[COMMAND_EXECUTION]: The skill uses automated shell commands to manage implementation environments and verify results.
Evidence: Phase 2 and 5 use mkdir and git branch -D for directory and branch lifecycle management.
Evidence: Phase 3 utilizes git worktree add to create isolated environments for parallel implementations.
Evidence: Phase 4 and 5 perform automated diffing and verification via diff -r, npm test, and npm run build.

cookoff