skills.rest
Chachamaru127
Chachamaru127COMMUNITY
Explore all skills by this author →
𝕏@vibecoder_japanGitHub

cursor-ask

Delegates read-only questions and design reviews to cursor-agent in locked ask mode.

3.1k|302|Updated Dec 12, 2025
Search Tags:#delegation#design review#read-only#second opinion#cursor-agent#Code review#ask mode
One-click install
npx skills add https://github.com/Chachamaru127/claude-code-harness --skill cursor-ask
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: cursor-ask
Source: https://github.com/Chachamaru127/claude-code-harness/tree/main/codex/.codex/skills/cursor-ask
Command: npx skills add https://github.com/Chachamaru127/claude-code-harness --skill cursor-ask

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Getting a second opinion from cursor-agent (Composer) normally requires worktree isolation, cherry-picking, and diff review to stay safe. This Skill removes that overhead by locking cursor-agent into ask mode, where it cannot write files or execute commands, so you can ask questions, run investigations, and get adversarial design feedback with zero containment setup.

Core Features & Use Cases

  • Read-Only Delegation: Passes your question to cursor-agent via cursor-companion.sh with --mode ask enforced, so no writes, no worktree, and no Lead diff review are needed.
  • Adversarial Sanity Checks: Ask Composer to find weaknesses in a design, list oversights in a diff, or challenge optimistic assumptions before you commit to an approach.
  • Host-Summarized Output: The host agent condenses cursor's raw output into a 3-5 line summary with conclusion, rationale, caveats, and next steps instead of pasting untrusted output verbatim.
  • Use Case: Before merging a PR, ask "cursor:ask what is the single biggest weakness of this PR?" and receive a summarized second opinion without any risk of the delegate modifying your codebase.

Quick Start

Ask cursor for a second opinion by invoking the skill with a question such as: cursor:ask "review this design decision from Composer's perspective and list three oversights".

Frequently Asked Questions about cursor-ask

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I ask cursor-agent a question without giving it write access?▼

Invoke cursor:ask with your question as the argument. The skill calls cursor-companion.sh in task mode, which automatically applies --mode ask, a hard read-only stop that prevents cursor-agent from writing files or running commands.

What is the difference between cursor:ask and cursor:do?▼

cursor:ask is read-only for questions, investigation, and design feedback, requiring no worktree or diff review. cursor:do handles write tasks like implementation and refactoring, and therefore needs full containment with worktree isolation, Lead diff review, and cherry-pick.

Can cursor-agent leak secrets when used in read-only ask mode?▼

Yes, read access can still expose secrets to cursor's inference. You must configure .cursorignore to exclude .env, *.pem, *.key, .ssh, .aws, and .git, and treat cursor output as untrusted since allowlists are best-effort, not a security guarantee.

Why does cursor:ask fail with a plugin root not found error?▼

The skill resolves the harness plugin root by checking for scripts/cursor-companion.sh plus a plugin.json marker, searching CLAUDE_PLUGIN_ROOT, the skill directory ancestry, and plugin cache paths. The error means none of these locations contained a valid installation.

When should I not use cursor:ask?▼

Do not use it for implementation, refactoring, file edits, or commit and push work, since ask mode cannot write. Use cursor:do or breezing --cursor instead for any task requiring write access.

Related Skills

Explore More Software Engineering Skills→
tylern91tylern91

ai-coding-agent-guardrails

Enforce permission boundaries, secret scanning, and audit trails for AI coding agents.

Community
Advanced
mei28mei28
3

antigravity-implement

Delegate implementation to the Antigravity CLI in headless mode and summarize the resulting diff.

Community
Intermediate
filipnyquistfilipnyquist

codex

Review code diffs with OpenAI Codex CLI and output a gate verdict.

Community
Advanced
whynpc9whynpc9

cursor-cli-review

Coordinate Cursor CLI agents for code review and planning in non-interactive mode.

Community
Advanced