Claude Skills

Claude Skills Bridge

Treat Claude Code skills as first-class Cursor skills. Discover them on disk, read the matching SKILL.md, and follow it immediately.

Skill locations (search in this order)

1. Project skills — <workspace>/.claude/skills/*/SKILL.md
2. Personal skills — ~/.claude/skills/*/SKILL.md
3. Installed Claude plugins — ~/.claude/plugins/cache/*/*/skills/*/SKILL.md
   (and any newer plugin skill paths under ~/.claude/plugins/)

If a skill name exists in more than one place, prefer project over personal over plugin.

When this applies

• User runs a Claude-style command: /skill-name …
• User names a skill: "use my-deploy-skill", "run code-review on …"
• Task matches a Claude skill description already on disk
• User asks what Claude skills are available

If Cursor already lists a matching path under available_skills, read that path. Otherwise discover via the locations above.

Workflow

1. Discover (only if needed)

List skill dirs / SKILL.md files from the locations above. For a quick inventory:

# Project
find "$WORKSPACE/.claude/skills" -name 'SKILL.md' 2>/dev/null

# Personal
find "$HOME/.claude/skills" -name 'SKILL.md' 2>/dev/null

# Plugins
find "$HOME/.claude/plugins" -path '*/skills/*/SKILL.md' 2>/dev/null

Read only the YAML name + description (and optional argument-hint) from each SKILL.md when listing. Do not load full skill bodies until one is selected.

2. Select

Match by:

• Slash command / folder name (my-skill.local → my-skill)
• Frontmatter name
• Description trigger phrases

If multiple skills match and the intent is ambiguous, ask which one. If none match, say so and offer the inventory.

3. Execute

1. Read the full SKILL.md with the Read tool before doing the task. Do not rely on memory or a partial summary.
2. Follow it as the source of truth — same steps, tools, output format, and constraints.
3. Resolve ${CLAUDE_SKILL_DIR} to the directory that contains that skill's SKILL.md (not the repo root). Example: for
   ~/.claude/plugins/cache/example-tools/example/0.1.0/skills/example-skill/SKILL.md
   → CLAUDE_SKILL_DIR = that example-skill directory. Rewrite script paths accordingly before running them.
4. Load linked files the skill references (e.g. reference.md, scripts) only as needed.
5. Adapt tools, not the workflow — if the skill names Claude-only tooling, use the closest Cursor equivalent (Shell, MCP, browser tools, etc.) while keeping the same procedure and success criteria.
6. Mention which Claude skill you are following (name + path) once at the start of the work.

4. List-only requests

When the user asks what Claude skills exist, return a compact table:

| Name | Source | Description (short) | Path |

Sources: project | personal | plugin:<plugin-id>.

Conventions

• Folder suffixes like .local are still valid skills; match them by base name or full folder name.
• Claude plugin skills may declare allowed-tools with Bash/node scripts — run those scripts via Shell after resolving CLAUDE_SKILL_DIR.
• Do not copy Claude skills into ~/.cursor/skills/ unless the user asks; bridge by reading in place.
• Do not create duplicate project docs about this bridge.

Examples

User: /my-skill some-arg
→ Read <workspace>/.claude/skills/my-skill/SKILL.md (or my-skill.local) and follow it.

User: "Use the example-skill from my Claude plugin"
→ Read that skill from ~/.claude/plugins/.../skills/example-skill/SKILL.md and follow it.

User: "What Claude skills do I have?"
→ Inventory project + personal + plugin skills; do not execute any.