The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a process that ingests external project data, such as instruction files (e.g., AGENTS.md) and design documents, to generate new specification files. This constitutes an indirect prompt injection surface. The risk is managed by a structured multi-phase workflow and the use of isolated subagents for code exploration and final document review.
Ingestion points: The agent is instructed to read project-level instruction files, approved proposals, and existing designs during Phase 1.1.
Boundary markers: The workflow uses mandatory template sections and HTML comments defined in Phase 0 as structural delimiters.
Capability inventory: The skill includes instructions for executing openspec CLI commands and writing final specification files to an outputPath in Phase 3.1.
Sanitization: The process includes mandatory interactive analysis with user confirmation gates and a separate subagent for artifact compliance review in Phase 3.4.
[COMMAND_EXECUTION]: The skill uses shell commands to interact with the openspec CLI tool for resolving templates and checking artifact status. These commands are essential to the skill's purpose and are executed within the local environment context.

openspec-plus-spec