The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the openspec CLI to manage workflows, including creating changes (openspec new change), checking status (openspec status), and retrieving instructions (openspec instructions). These tools are explicitly listed in the allowed-tools frontmatter and are scoped to the project's planning environment.
[INDIRECT_PROMPT_INJECTION]: The workflow involves reading external content from openspec instructions --json, which provides the agent with instructions, rules, and templates. While this is an ingestion point for untrusted data, the skill includes a 'Planning boundary' guardrail that explicitly instructs the agent not to modify project code and to stop after artifacts are created, reducing the risk of unauthorized actions.
[DATA_EXPOSURE]: The skill reads and writes to specific paths resolved by the CLI (e.g., planningHome, changeRoot). This is limited to the project's documentation and planning artifacts and does not appear to access sensitive system files or credentials.
[DYNAMIC_EXECUTION]: The skill may invoke other skills or commands if the instruction field from the CLI tool delegates the creation of an artifact. This behavior is mediated by the openspec CLI tool, which is the primary interface for this skill's functionality.

openspec-propose