The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the openspec CLI to perform actions such as creating changes, checking status, and retrieving instructions. It passes user-supplied input (e.g., change names) to these commands. The skill includes a mitigation strategy by instructing the agent to convert user input into a sanitized kebab-case format before execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from two main sources: user descriptions of requested changes and existing project artifacts read from the disk. This creates a surface for indirect prompt injection where malicious instructions embedded in project files or user requests could attempt to influence the agent's output. The skill mitigates this by using specific templates and status checks to constrain the agent's behavior.
[DATA_EXPOSURE]: The skill reads existing artifact files to provide context for generating new ones. While this involves reading local files, the operations are scoped to the project directory resolved by the CLI and no network exfiltration patterns were detected.

openspec-propose