The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to install the Cursor Agent CLI by fetching an installation script from https://cursor.com/install and piping it directly to the bash shell.
[COMMAND_EXECUTION]: The skill makes extensive use of shell tools to manage temporary workspace directories in /tmp/, manipulate plan files using cat and tee, and execute multiple external AI CLI utilities (agent, claude, codex) with arguments constructed from local file content.
[DATA_EXFILTRATION]: Technical plans and codebase snippets are transmitted to external third-party AI providers (Anthropic, OpenAI, Cursor) via their respective command-line interfaces for the purpose of conducting multi-model consensus reviews.
[PROMPT_INJECTION]:
The skill defines a sub-agent workflow using templates (assets/reviewer-prompt-template.md) that employ role-play instructions, directing the AI to act as a senior engineer.
Indirect Prompt Injection Surface:
Ingestion points: The skill reads and processes plan content from generated files stored in the temporary planning directory.
Boundary markers: Employs a specific markdown header (# Proposal to review) to delineate instructions from the data being reviewed in the prompt template.
Capability inventory: The skill possesses capabilities to execute shell commands, write to the file system, and invoke external network-connected CLIs.
Sanitization: There is no evidence of explicit sanitization or escaping of the plan content before it is interpolated into prompts for the reviewer models.

multi-model-planning