The Agent Skills Directory

[SAFE]: The skill defines a standard software engineering lifecycle for sub-agents, requiring them to read specifications, designs, and existing code before performing modifications.
[COMMAND_EXECUTION]: The skill instructs the agent to detect and execute test runners (e.g., npm test, pytest, make test) based on project configuration files like package.json or pyproject.toml. This is expected behavior for a coding assistant to verify implementation correctness.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data including task lists, design documents, and existing project source code.
Ingestion points: Reads files from sdd/ directory, tasks.md, config.yaml, and existing source files within the workspace.
Boundary markers: The skill uses structured sections and specific task IDs to maintain context, though explicit "ignore instructions" delimiters for external content are not defined.
Capability inventory: The skill can write to the filesystem, update memory state via mem_update, and execute shell commands for testing.
Sanitization: No specific sanitization of external text is defined, but the skill's logic is constrained by the requirement to pass tests and match existing project patterns.

sdd-apply