The Agent Skills Directory

[SAFE]: The skill implements a transparent and highly structured development workflow. Each phase is clearly defined in its own file (e.g., phases/requirements.md, phases/design.md) and uses templates to maintain consistency.
[COMMAND_EXECUTION]: During the implementation phase (phases/impl.md), the skill identifies and executes local project verification commands (such as linting, type-checking, and running tests). These commands are discovered from standard project files like package.json, Makefile, or pyproject.toml, which is standard behavior for development-oriented AI agents.
[EXTERNAL_DOWNLOADS]: The discovery guidelines (rules/discovery.md) recommend using documentation lookup, web search, and fetch tools to verify library APIs and third-party service contracts. This is conducted for research purposes to ensure design accuracy and is consistent with the skill's stated purpose of technical planning.
[PROMPT_INJECTION]: The skill uses subcommands (/spec init, /spec requirements, etc.) and a routing mechanism in SKILL.md to guide the agent's behavior. The instructions are professional and focused on the development process without any attempts to bypass safety filters or override system-level instructions.
[SAFE]: The skill includes a 'Self-Review via Subagent' pattern (rules/self-review.md) which uses a fresh agent context to validate generated artifacts (like design or requirements documents) against specific quality criteria before they are presented to the user. This serves as a robustness and quality control measure.

spec