The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute the Cursor CLI agent command to perform various tasks like building features, refactoring, and code review. This is the primary and intended function of the skill.
[COMMAND_EXECUTION]: The instructions mention several security-sensitive flags for the Cursor CLI, such as --trust (to bypass workspace trust prompts), --force, --yolo, and --approve-mcps. The skill provides appropriate cautionary guidance, labeling these as "intentional overrides" and instructing the agent not to use them silently without explicit user intent.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates processing untrusted data (local codebases and external files) by passing task descriptions to the Cursor agent.
Ingestion points: Local project files and user-supplied task strings in SKILL.md examples.
Boundary markers: Example commands use double quotes to encapsulate task descriptions.
Capability inventory: The agent command has significant capabilities including file system modification and potentially tool execution (via MCP).
Sanitization: The skill does not implement specific sanitization, relying instead on the user's explicit request and the underlying tool's internal guardrails.

cursor-coding-agent