More room for your next idea. Membership from $9.99/month.Take a look

YOUR AGENT. YOUR WAY OF WORKING.

Cursor skills for your next project

Bring reusable creative workflows into your editor. Choose a skill by its output, inspect the examples and supporting files, then adapt the workflow to the project you have open in Cursor.

Browse released packages Installation guide →

What are Cursor skills?

An Agent Skill is a folder centered on SKILL.md, with instructions and any supporting templates, examples or scripts. Install it in the directory your agent reads, then give the agent your task. Skillrove provides the package; your agent runs the workflow in your own environment.

Available here
84 released packages
Free packages
45 · verified account required
Premium packages
39 · purchase or active membership

Installation locations

Project locations
.cursor/skills/SKILL_SLUG/SKILL.md .agents/skills/SKILL_SLUG/SKILL.md
Personal locations on this machine
~/.cursor/skills/SKILL_SLUG/SKILL.md ~/.agents/skills/SKILL_SLUG/SKILL.md

Inspect discovered skills in Customize

Open Customize → Skills and look for the package. Cursor discovers skill folders at startup, including scoped folders in repository subdirectories. Keep the containing folder name consistent with the SKILL.md name and check any paths filter against the files you are working on.

Select a skill from / in Agent chat

Type / in Agent chat and select debug-evidence-kit. Selection attaches the skill to a message; a Custom Mode keeps it in the session. disable-model-invocation: true limits selection to explicit invocation, rather than matching the task automatically.

/debug-evidence-kit Run the included isolated failing/fixed/passing exercise in a new folder, then explain the source correction and verification limits.
Cursor documentation ↗

Host documentation checked on 14 October 2026. Verify the controls in the product and version you actually use.

A reproducible first run

Read the package and inspect its real output before trying the included example in your own agent.

Debug Evidence Kit · v1.0.0 →

A reproducible fix and proof handoff

  • report/index.html
  • report/case.json
  • report/fix-handoff.md
  • report/delivery-manifest.json
  • private evidence/before.json + after.json
Open the real output →

Use Python 3.10+ standard library for the isolated fictional fixture. The public case records real failing, corrected and passing test executions. Inspect the source diff and unchanged checks; a passing fixture does not certify every feature in your own project.

Run from the installed package directory into a fresh output folder.

python3 scripts/run-example.py --output ./sample-run
Read the task guide →

CHOOSE BY THE RESULT

Skills for Cursor

These published packages list Cursor as an installation host. Check each skill’s setup and dependencies; labels do not certify every workflow has been tested in every agent.

Filter the full library ↗

ZIP or CLI: choose how to download

The ZIP route needs no Node.js installer. Extract the complete package into the chosen host folder. Node.js 24+ is required by the standalone Skillrove CLI; the workflow itself may have additional runtime requirements.

After browser authorization, install into an unused project folder. The CLI target is explicit; this is not a host-discovery check or a script execution.

node skillrove-cli.mjs login
node skillrove-cli.mjs install debug-evidence-kit .cursor/skills/debug-evidence-kit

Read the installation instructions →

Cursor discovery and remote-scope details →

Check package access →

Local skills and Cloud sync have different reach

Personal Cloud sync is opt-in and covers ~/.cursor/skills only. ~/.agents/skills and unsynced local files are not copied to Cloud Agents or remote SSH sessions. For a repository workflow, make sure the relevant project skill and dependencies exist in the remote checkout.

Cursor can discover compatibility directories from other agents, but that does not establish equivalent interpretation of their tool permissions or frontmatter. Review local commands, allowed files and output checks for the editor session you actually use.

Compare host compatibility →

Find the stage that failed

Separate download access, discovery, invocation, dependencies and exported output. A visible skill or browser preview is only one stage; use the linked troubleshooting guide to inspect the first failing check.