The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as project proposals, design documents, and user responses to generate specifications. It mitigates risks by implementing a strict four-phase workflow (Phase 0-3) that prevents immediate writing and enforces an interactive analysis phase. The capability inventory includes file writing to specific project paths and internal CLI execution, which are standard for documentation tools.
[COMMAND_EXECUTION]: The skill executes the openspec CLI tool to resolve schemas and templates. This tool is part of the vendor's (Elastic) specification framework and is used for project-specific configuration retrieval. No arbitrary or unsanitized shell execution was detected.
[DYNAMIC_EXECUTION]: The skill dispatches subagents (e.g., general-purpose for review and explore for behavior analysis) to perform specialized tasks. These dispatches are constrained by strict prompts that limit the subagents to describing observable behaviors without accessing implementation details or executing code in the root context.
[PRIVILEGE_ESCALATION]: The skill does not request or utilize elevated permissions, sudo, or system-level modifications. File operations are restricted to the outputPath defined by the project's internal configuration.

openspec-plus-spec