Keep · infra SKILL.md
Package a Cursor SKILL.md (discover → activate → scripts)
WhenUse when a workflow needs a folder with SKILL.md, then references/ and scripts/ on demand.
StopNever bloat the body if the Agent skips the skill — fix description first. Never leave Cloud Agents without a repo copy.
Steps
- Fast path: in Agent chat type /create-skill, describe the workflow, save under .cursor/skills/<name>/. name must match the folder (lowercase, digits, hyphens).
- Manual: mkdir .cursor/skills/your-skill-name/ and write SKILL.md with YAML name + description. Description is the router — name the user phrasings and the output.
- Project skills: .cursor/skills/ or .agents/skills/ (commit them). Personal: ~/.cursor/skills/ or ~/.agents/skills/ (not copied to Cloud Agents / remote workers — bake those into the repo or the worker image).
- Optional paths: globs so the skill only surfaces on matching files. Nested .cursor/skills/ under apps/web/ auto-scopes to that subtree.
- Put repeatable commands in scripts/ and tell the body when to run them and how to roll back. Keep SKILL.md short; dump long reference into references/.
- Acceptance test: run a real task. If the Agent skips the skill, fix description before bloating the body. Also loads .claude/skills/ and .codex/skills/ for compatibility.
Load.cursor/skills/ or .agents/skills/. Optional paths globs.
SourceCursor docs
--- name: Package a Cursor SKILL.md (discover → activate → scripts) description: Use when a workflow needs a folder with SKILL.md, then references/ and scripts/ on demand. --- # Package a Cursor SKILL.md (discover → activate → scripts) When: Use when a workflow needs a folder with SKILL.md, then references/ and scripts/ on demand. Stop: Never bloat the body if the Agent skips the skill — fix description first. Never leave Cloud Agents without a repo copy. 1. Fast path: in Agent chat type /create-skill, describe the workflow, save under .cursor/skills/<name>/. name must match the folder (lowercase, digits, hyphens). 2. Manual: mkdir .cursor/skills/your-skill-name/ and write SKILL.md with YAML name + description. Description is the router — name the user phrasings and the output. 3. Project skills: .cursor/skills/ or .agents/skills/ (commit them). Personal: ~/.cursor/skills/ or ~/.agents/skills/ (not copied to Cloud Agents / remote workers — bake those into the repo or the worker image). 4. Optional paths: globs so the skill only surfaces on matching files. Nested .cursor/skills/ under apps/web/ auto-scopes to that subtree. 5. Put repeatable commands in scripts/ and tell the body when to run them and how to roll back. Keep SKILL.md short; dump long reference into references/. 6. Acceptance test: run a real task. If the Agent skips the skill, fix description before bloating the body. Also loads .claude/skills/ and .codex/skills/ for compatibility. Load: .cursor/skills/ or .agents/skills/. Optional paths globs.