The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from project artifacts and CLI outputs (specifically the context and operationGuidance fields). This data is interpolated into the agent's context to influence implementation facts and conventions.
Ingestion points: Reads files listed in contextFiles and parses JSON output from openspec instructions apply.
Boundary markers: The skill includes instructions to treat specific fields as separate from state and to not copy them into implementation files.
Capability inventory: Uses Bash(openspec:*) and performs file system writes to implement tasks.
Sanitization: The skill advises the agent to report conflicts and preserve CLI-controlled values, though it does not provide strict schema validation for the injected context content.
[COMMAND_EXECUTION]: The skill generates and executes shell commands for the openspec CLI by dynamically interpolating user-provided inputs such as change name and store id into command strings. This presents a potential command injection surface if inputs are not properly sanitized by the underlying platform.

openspec-apply-change