The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands such as openspec --help and fd to verify the environment and project state. These operations are restricted to discovery and validation of the local workspace.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external PRD files (docs/prd/{slug}.md). While these files represent an untrusted input surface, the skill mitigates risk by requiring the agent to present a breakdown to the user and wait for explicit approval before creating any specification files.
[SAFE]: The instructions emphasize a 'human-in-the-loop' design, ensuring that the AI does not implement changes or write files without iterative user confirmation. It also specifically avoids hardcoding CLI verbs, reducing the risk of executing unintended or malicious commands if the environment is compromised.

z-prd-to-openspec