The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads untrusted external data from PRD files (docs/prd/) and issue trackers. These inputs influence the agent's code implementation and command execution phases.
Ingestion points: Reads product requirement documents (PRDs) and issue bodies from the issue tracker as defined in Step 2 of SKILL.md and REFERENCE.md.
Boundary markers: Absent. The skill does not explicitly instruct the agent to ignore potentially malicious instructions embedded within the requirements or documents.
Capability inventory: The skill allows for arbitrary file writes during implementation (Step 3c), performs git commits (Step 3f), and executes shell commands for typechecking and testing (Steps 3d and 3e).
Sanitization: Absent. Data from requirement documents is used directly to guide implementation and file modification without validation or escaping.
[COMMAND_EXECUTION]: The skill executes various local developer tools based on project configuration. While these are standard for a development agent, they represent a shell execution surface triggered by the content of the project being worked on.
Evidence: Step 3d and 3e in REFERENCE.md specify running commands like tsc --noEmit, mypy ., cargo check, make typecheck, npm run typecheck, npm test, and pytest.

implement