The Agent Skills Directory

[SAFE]: The skill is designed for architectural planning and implementation roadmap generation. It includes explicit constraints against writing production code, inspecting sensitive configuration files like .gitignore, or performing Git commits without user intervention.
[DATA_EXPOSURE]: While the skill reads local design documents, PRDs, and repository metadata (commits, interfaces), it uses this information strictly for verification (SHA-256 checks) and planning. No network operations or access to sensitive credential stores are present.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from design documents and PRDs provided by the user. This creates a surface for indirect prompt injection if these documents contain instructions intended to derail the agent. However, since the skill's output is limited to a markdown implementation plan and it lacks execution capabilities (subprocess, network, etc.), the risk is minimal. Mandatory Evidence Chain:
Ingestion points: Reads design and PRD files from paths provided in current context (SKILL.md Stage 1).
Boundary markers: Uses a structured markdown template (assets/plan.md) for output.
Capability inventory: Limited to file read and local file write to specific documentation paths.
Sanitization: Relies on SHA-256 validation to ensure files match a previously validated state.

to-plan