The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes architecture documents to generate task lists, which introduces a surface for indirect prompt injection. Malicious content in source documents could influence the resulting implementation tasks.
Ingestion points: External architecture documents referenced during the planning process.
Boundary markers: The instructions do not mandate delimiters or safety warnings for separating source content from task instructions.
Capability inventory: The skill facilitates file system inventory via dark-factory list and generates commands for cargo, bun, and dark-factory.
Sanitization: No explicit sanitization or validation of content extracted from documents is required by the instructions.
[DYNAMIC_EXECUTION]: The skill generates a JSON task graph containing shell commands in the verification_steps field (e.g., cargo build, bun test). These commands are dynamically derived from external architecture documents and are intended for execution by autonomous agents.

planning-implementation