The Agent Skills Directory

[SAFE]: The skill operates within the project environment, reading documentation and writing new specifications. It accesses context files such as CLAUDE.md and ai-context/ to maintain architectural consistency and does not exfiltrate data or perform unauthorized actions.
[PROMPT_INJECTION]: The skill processes project files as input, creating an indirect prompt injection surface. ● Ingestion points: openspec/changes/<change-name>/proposal.md, ai-context/, and CLAUDE.md. ● Boundary markers: None present. ● Capability inventory: File system read/write operations within the project directory. ● Sanitization: No explicit content validation performed. This risk is mitigated by human-in-the-loop requirements for contradictions and the limited scope of the tool's output.

sdd-spec