The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates and manages executable code within the scripts/arch/checks/ directory. The template assets/arch-fitness/checks/example.mjs.template uses Node.js execSync to run shell commands that are derived from user-defined architectural criteria.
[COMMAND_EXECUTION]: The assets/arch-fitness/verify.mjs runner executes the generated check scripts as sub-processes using spawnSync, allowing for the programmatic evaluation of compliance rules.
[EXTERNAL_DOWNLOADS]: As detailed in references/dependencies.md and references/fitness-functions.md, the skill is authorized to identify, propose, and install external tools such as dependency-cruiser, ESLint, and Vitest using the project's native package managers (e.g., npm, pip, composer).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input to generate the content of ADRs and standards, which includes the logic for fitness functions. This creates a surface where malicious input could lead to unauthorized command execution.
Ingestion points: User descriptions for architectural decisions and requirements processed in SKILL.md.
Boundary markers: The templates do not utilize explicit boundary markers or delimiters to isolate user-provided data from code templates.
Capability inventory: The skill has access to shell execution (execSync/spawnSync) and package management tools.
Sanitization: There is no documented mechanism to sanitize or validate user input before it is interpolated into shell command strings within the generated scripts.

arch-manage