The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the openspec CLI tool via Bash to manage specification stores, check status, and validate changes. Tool access is limited to the openspec namespace as defined in the allowed-tools configuration.
[INDIRECT_PROMPT_INJECTION]: The skill processes "delta spec" files which are external markdown documents. This represents an attack surface where malicious instructions could be embedded in spec content to influence the agent during the merging process.
- Ingestion points: SKILL.md (Step 4a) reads delta spec files from paths provided by openspec status output.
- Boundary markers: Absent. The skill relies on specific markdown headers (e.g., ## ADDED Requirements) to structure data, but does not use explicit boundary markers to isolate data from instructions.
- Capability inventory: SKILL.md performs file writes (Step 4c/d), file/directory deletion (Step 4c), and executes openspec CLI commands (Steps 2, 4, 5).
- Sanitization: The skill performs semantic merging of markdown content based on headers but does not explicitly sanitize the text of requirements against prompt injection patterns.

openspec-sync-specs