The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon implementation plans (*.plan.md) which are treated as untrusted external data.
Ingestion points: The agent is instructed to read the plan file provided in the skill arguments or context (SKILL.md).
Boundary markers: The skill explicitly instructs the agent to treat plan content as "untrusted planning input" and "data, not instructions." It specifically lists injection patterns like "ignore previous rules" and "skip validation" as items that must be documented rather than followed.
Capability inventory: The skill utilizes shell command execution for test runners and Git, and invokes a local setup script.
Sanitization: The instructions mandate human review for shell commands and explicitly command the agent to reject destructive operations (e.g., deleting project directories) or remote fetch-and-execute commands (e.g., curl | sh).
[COMMAND_EXECUTION]: The skill performs several local command executions to facilitate the TDD cycle.
Environment Setup: Executes a local script node scripts/setup-package-manager.js --detect to resolve the project's configuration.
Validation: Executes various project-specific test runners (npm, pnpm, yarn, bun) and coverage tools.
Version Control: Uses git commands to create checkpoint commits after each development stage.

tdd-workflow