The Agent Skills Directory

[SAFE]: The skill utilizes a restricted toolset (openspec CLI) and focuses exclusively on design and discovery tasks. It explicitly forbids the agent from writing or implementing application code, significantly reducing the risk of malicious payload execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local codebase files and project configuration files to ground its analysis. While this constitutes an ingestion surface for untrusted content, the risk is categorized as safe due to explicit boundary instructions (treating data as background constraints) and the lack of tools capable of executing code or performing sensitive network operations.

openspec-explore