The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the openspec CLI tool to perform operations like creating changes, checking status, and retrieving artifact instructions. These commands are executed via the shell.
[INDIRECT_PROMPT_INJECTION]: The agent's behavior is guided by data retrieved at runtime from the openspec instructions command. This JSON output contains a instruction field that the agent is told to follow as authoritative guidance.
Ingestion points: The agent parses JSON output from shell commands executed in SKILL.md.
Boundary markers: The instructions include clear markers directing the agent to use context and rules fields as internal constraints only and to exclude them from the final generated artifacts.
Capability inventory: The skill has the capability to write files to disk and execute shell commands within the openspec tool scope.
Sanitization: There is no explicit sanitization or validation of the text provided in the tool's instruction or template fields before the agent processes them.
[DYNAMIC_EXECUTION]: The skill's logic allows the instruction field from the tool's output to delegate the creation of an artifact to another specific command or skill. While the execution environment is restricted to the openspec namespace in the frontmatter, this pattern allows for dynamic selection of executable paths based on external data.

openspec-propose