The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/runtime/common.mjs defines a utility function runChild that uses node:child_process.spawn. This function appears to be a helper for potential internal tooling or domain adapters, but it is not invoked by the skill's primary execution paths or integrity validation scripts.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and structure information from various project-level files, which constitutes a potential attack surface if those files are maliciously crafted.
Ingestion points: According to SKILL.md, the agent reads from .devflow/index.md, member 'Sketch' state files, and project briefs.
Boundary markers: The skill explicitly requires the use of YAML frontmatter headers in documentation and follows strict behavioral contracts defined in the references/ directory.
Capability inventory: The skill performs localized file system operations (read/write) within the project directory and executes integrity checks on its own artifacts using standard Node.js modules.
Sanitization: The instructions mandate verification against 'canonical' sources and 'landing conditions' to prevent the integration of unvalidated or contradictory information into the project definition.

product