Ashwin Skills — Claude Code & Cursor agent skills

License: MIT Claude Code Cursor

Agent skills for Claude Code, Cursor, and other coding agents — workflows I use every day to ship real product, not vibe coding.

Cross-platform session handoff · UI render verification · agent design with a singular outcome.

Born from failure modes switching between Claude Code and Cursor when plan limits hit mid-task.

Install (Claude Code, Cursor, Codex, …)

npx skills add freezingfunky/ashwin-skills

Or copy folders from skills/ into your agent skills directory:

Agent Skills path
Claude Code ~/.claude/skills/
Cursor ~/.cursor/skills/
Codex / others see your agent’s docs

Skills

Skill What it solves Keywords
dual-tool-handoff Claude Code ↔ Cursor handoff + fetch/hydrate from the other tool’s local store; Diff check warns if shipped ≠ tree claude code cursor handoff, fetch session, named session
render-check Prove a UI fix actually rendered (served HTML/CSS, not just source) fast refresh, next.js, tailwind, stale build
agent-design Design LLM agents with a decision map + Task · Context · Memory · Eval agent design, eval, singular outcome

Why these skills exist

Sessions don’t sync between Claude Code and Cursor

Hit a Claude Max / plan limit mid-refactor, switch to Cursor — blank chat, context gone. Pasting is lossy.

Fix: /dual-tool-handoff — named handoff file in the repo; resume the same session name on the other tool. If you forgot to hand off, fetch/hydrate from the other tool’s local store (confirm first). Switch-only (not live sync). Includes Diff check: git paths vs agent shipped narrative (warn ≠ block; Diff ≠ why).

The agent said the UI was fixed. The screen didn’t change.

Source edited, typecheck green, Fast Refresh stale — source ≠ served HTML/CSS.

Fix: /render-check — prove the live markup and compiled CSS changed before calling it done.

The agent can do everything — so it does nothing reliably

Flaky agents often wrap multiple outcomes and skip eval.

Fix: /agent-design — decision-map forces a singular outcome (or splits + optional parallel fan-out); then lock the four pillars.

Quick start — dual-tool handoff

1. In Claude Code: “hand off to Cursor”
2. Pick a session name (or continue an existing one)
3. Skill writes .scratch/handoffs/<slug>.md
4. In Cursor: “resume <that name>”
   — or, if you forgot step 1: “fetch <name> from Claude”

The session isn’t the source of truth. The repo file is.

Repo layout

skills/
  dual-tool-handoff/   # Claude Code ↔ Cursor handoff
  render-check/        # verify UI actually rendered
  agent-design/        # design agents on purpose

Related searches

Claude Code skills · Cursor skills · agent skills · AI coding agent workflows · cross-platform handoff · Claude Max limit · Fast Refresh staleness · LLM agent design

License

MIT — see LICENSE.