Cursor Specialist Skill
Purpose: Expert guidance on Cursor IDE — rules, agents, MCP, configuration. Reference:
~/toolkit/docs/cursor-manual.md— state of the art snapshot. Invoked by: user, prompt-factory, architect, or any agent working with Cursor features.
When to Use
- Creating or modifying Cursor rules (
.cursor/rules/*.md, frontmatter, application modes) - Configuring MCP servers in Cursor (
.cursor/mcp.json) - Setting up custom modes or agent configurations
- Deploying toolkit rules to a project (via
cursor-rules-link.sh) - Comparing Cursor vs Claude Code capabilities
- Debugging Cursor agent behavior, tab completion, context issues
When NOT to Use
- Claude Code features (use
claude-code-specialist) - General prompt engineering (use
prompt-certification) - Model selection (use
anthropic-models-specialist)
Process
- Freshness check: Run
~/toolkit/scripts/tool-freshness.sh cursor. If stale (exit 1), do a targeted refresh of the section relevant to your current task before proceeding — check the Sources in the manual for that section's official docs. Update the manual andlast_refreshedinconfig/tools.jsonafter refreshing. - Read
~/toolkit/docs/cursor-manual.mdfor current state of the art - Verify against official docs before making changes (see Sources in the manual)
- Check detection rules below against the current state
- Apply changes following the manual's documented patterns
- Update the manual if something changed
Detection Rules
| # | Detect | Severity | Fix |
|---|---|---|---|
| 1 | .cursorrules file present |
WARNING | Migrate to .cursor/rules/*.md — .cursorrules is deprecated |
| 2 | Rule with alwaysApply: true and globs set |
WARNING | Redundant: alwaysApply ignores globs. Remove one |
| 3 | Rule expecting to affect Tab completion | BLOCKER | Rules do NOT affect Tab — only Agent and Inline Edit |
| 4 | User Rules expecting Inline Edit (Cmd+K) | WARNING | User Rules only apply in Chat Agent, not Inline Edit |
| 5 | MCP using SSE transport | WARNING | SSE deprecated. Switch to Streamable HTTP |
| 6 | .cursorignore treated as security boundary |
BLOCKER | Best-effort only. Terminal and MCP tools bypass it |
| 7 | Single rule file > 500 lines | WARNING | Split into multiple composable rules |
| 8 | Auto mode rule without description |
BLOCKER | Intelligent mode needs description to decide relevance |
Refresh Mode
Last refreshed: 2026-03-05
To update the knowledge snapshot:
- Read
~/toolkit/docs/cursor-manual.md— locate the Sources section at the bottom - Visit each URL in the Sources tables using WebSearch and WebFetch:
- Official Docs: check rules, modes, MCP, models, tab docs for changes
- Changelog: scan for new versions, features, breaking changes
- Community: check Cursor Blog and Forum for patterns
- Compare findings with current snapshot content:
- New version? Update version, "Cambios recientes" table
- New agent modes? Update Agent / Modos section
- New rule frontmatter fields? Update Rules section
- MCP changes? Update MCP section
- Breaking changes? Update detection rules
- New pain points? Update Known Pain Points table
- Update changed sections in
~/toolkit/docs/cursor-manual.md - Update the "Last refreshed" date in this skill and in the manual's Snapshot field
- If detection rules changed, update the table above
Search queries for refresh:
site:cursor.com/changelog(new versions)site:cursor.com/docs(official doc updates)cursor IDE new features 2026(community coverage)cursor rules format changes 2026(rules-specific)
Quick Reference
| I want to... | Where to look |
|---|---|
| Create a rule | Manual -> Rules -> Formato MDC |
| Configure MCP | Manual -> MCP -> Formato JSON |
| Set up custom mode | Manual -> Agent / Modos -> Custom Modes |
| Deploy toolkit rules | scripts/cursor-rules-link.sh |
| Ignore files from AI | Manual -> .cursorignore |
| Compare with Claude Code | Manual -> Features + docs/claude-code-manual.md |
| Check latest changes | Manual -> Cambios recientes |
Source: brianmontero18/astral — distributed by TomeVault.