⏳ This skill is pending AI review.

Scores will appear once the review pipeline completes.

version unknown

cursor-engineering-mode

@olddonkey⭐ 16 stars

Own goal-first engineering outcomes: investigate, find root cause, design, plan, then delegate implementation unit-by-unit to cursor-implementation-loop, this plugin''s kernel skill, with verification and honest reporting. Use to fix, build, or own an outcome end to end; investigate-and-fix; "make X work"; or plan-then-execute in one request. A request with an approved plan or equivalently precise spec that needs nothing beyond implementation belongs directly to cursor-implementation-loop. Requires this plugin''s cursor-implementation-loop skill; they ship together.

Choose how to use this skill

You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.

1. Native installer

This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.

Do not guess an installer command or replace an existing version without reviewing the diff.

2. Complete package recommended

Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.

No complete ProSkills package is published for this listing yet.

3. Prompt-only

Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.

Need only the instruction file?

Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.

No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.

—/10

// RATINGS

⭐GitHub Stars
⭐⭐ 16 on GitHubGitHub ↗

Growing

🟢ProSkills Score
—
📍

Not yet listed on ClawHub or SkillsMP

// README

Olddonkey Skills

Open-source Agent Skills distilled from workflows that ran, broke, and got fixed.

License: MIT Spec

English · 简体中文


Three Claude Code skills so far, plus a Cursor Plugin that ships two engineering skills:

  • implementation-loop — delegate implementation to Codex, grok, or cursor-agent without delegating judgment: Claude reviews the real diff, runs the full test gate, and ships only what it would sign its name to. Those three are the agent-driven backend dial; Claude Code is also shipped as a fourth backend module for coordinator use.
  • engineering-mode — goal-first engineering ownership: investigate, design, and plan, then drive implementation-loop unit by unit.
  • cursor-implementation-loop — Cursor Plugin with the plan-first implementation loop and the goal-first cursor-engineering-mode wrapper; the parent agent reviews, gates, and publishes while an implementer subagent writes code.
  • web-slides — turn material or outlines into click-driven 16:9 HTML slide decks for live presenting, with 24 built-in themes and a presenter view that keeps speaker notes off the shared screen.

Installation

Marketplace (recommended)

Inside Claude Code, add the marketplace once, then install the skills you want:

/plugin marketplace add olddonkey/olddonkey-skills
/plugin install implementation-loop@olddonkey-skills
/plugin install engineering-mode@olddonkey-skills
/plugin install web-slides@olddonkey-skills
/reload-plugins

These plugins were renamed by dropping the codex- prefix when the loop gained Codex, grok, and cursor-agent backends. A Claude Code coordinator backend module is also available. If the former Codex-prefixed plugin IDs are installed, uninstall them and install implementation-loop and engineering-mode instead.

The implementation-loop adapter scripts moved from scripts/ into backends/<name>/ for the 0.4.x release. The old executable paths forward for that release so existing harnesses keep working; update external callers to the new module paths before the shims are removed in release 0.5.0.

engineering-mode requires implementation-loop; the Codex backend needs an authenticated codex CLI and no additional Claude Code plugin — see Setup below. web-slides needs nothing beyond Node.js for the generated slide project.

Manual

Copy into your personal skills directory:

git clone https://github.com/olddonkey/olddonkey-skills /tmp/olddonkey-skills
mkdir -p ~/.claude/skills
cp -R /tmp/olddonkey-skills/skills/<skill-name> ~/.claude/skills/

Or symlink a clone for pull-to-update:

git clone https://github.com/olddonkey/olddonkey-skills ~/Documents/olddonkey-skills
mkdir -p ~/.claude/skills
ln -s ~/Documents/olddonkey-skills/skills/<skill-name> ~/.claude/skills/<skill-name>

If you created ~/.claude/skills for the first time during an active Claude Code session, restart the session so the new top-level directory is discovered. If your agent does not follow symlinks in its skills directory, use the copy option and re-copy after git pull.

Cursor Plugin

For Cursor, the one-line installer copies the standalone skill and both agents into Cursor's discovered directories. It works on all current builds and on a personal or temporary/company machine — it only writes under your home directory:

One-line install (bare-copy default):

curl -fsSL https://raw.githubusercontent.com/olddonkey/olddonkey-skills/main/install-cursor.sh | bash

--copy remains accepted as an explicit alias. For older Cursor builds that scan plugins/local, opt into the managed-checkout symlink with --link:

curl -fsSL https://raw.githubusercontent.com/olddonkey/olddonkey-skills/main/install-cursor.sh | bash -s -- --link

Want true plugin form? In Cursor's Customize → Plugins page, press "+ Add" and select $OLDDONKEY_SKILLS_DIR (default: ~/olddonkey-skills); that checkout root contains .cursor-plugin/marketplace.json, which registers the cursor-implementation-loop plugin. Then remove the standalone copies to avoid double-loading.

Alternatively, install the local-plugin symlink manually:

git clone https://github.com/olddonkey/olddonkey-skills
mkdir -p ~/.cursor/plugins/local
ln -s "$(pwd)/olddonkey-skills/cursor-implementation-loop" \
      ~/.cursor/plugins/local/cursor-implementation-loop
# restart Cursor or run "Developer: Reload Window"

Verify once:

bash ~/.cursor/plugins/local/cursor-implementation-loop/skills/cursor-implementation-loop/scripts/gate-selftest.sh
# expect: selftest: PASS (207 checks)

To reproduce the default install manually, copy bare files — both steps are required:

mkdir -p ~/.cursor/skills ~/.cursor/agents
cp -R olddonkey-skills/cursor-implementation-loop/skills/cursor-implementation-loop \
      ~/.cursor/skills/
cp olddonkey-skills/cursor-implementation-loop/agents/*.md ~/.cursor/agents/

To uninstall a symlink install: rm ~/.cursor/plugins/local/cursor-implementation-loop and delete the clone. Team marketplaces (Teams/Enterprise) can also import this repository: Dashboard → Plugins → Import from Repo. Manifest: .cursor-plugin/marketplace.json.

Plan-first vs goal-first

Plan-first (unchanged): approve a plan, then invoke the implementation loop directly: Use implementation-loop to implement PLAN.md unit by unit.

Goal-first: give engineering mode an outcome; it investigates the root cause, chooses a design, writes the plan, then drives that same loop. Codex: Use engineering-mode to fix duplicate fulfillment caused by webhook replay. Cursor: Use cursor-engineering-mode to fix duplicate fulfillment caused by webhook replay.

Codex prerequisites: engineering-mode → implementation-loop → authenticated codex CLI. On Cursor, cursor-engineering-mode ships inside the cursor-implementation-loop plugin, so there is no separate install; existing installs receive it on update by rerunning the one-line installer or running git pull in the managed checkout.


implementation-loop

The agent-driven loop delegates to Codex, grok, or cursor-agent. A coordinator can also use the Claude Code backend module.

The new loop coordinator can draft a unit spec with a judge agent, record the engineer's approval, and review an existing working tree diff. In this unit it does not run an implementer, gate, commit, or publish.

The implementer implements and runs focused tests. Claude reviews the real diff, runs the full gate, and ships only what it would sign its name to.

The agent-driven backend dial offers Codex, grok, and cursor-agent. The default is Codex (setup below); the same loop and the same review-and-gate discipline apply whichever backend implements. Each backend's git and publication boundary works differently and is documented in its own runtime reference — read the selected backend's before its first dispatch:

  • Claude Code (coordinator module) — git-less copies, a fixed tool list, the CLI OS sandbox, and a checked patch. It is not a choice of the agent-driven loop’s backend dial. backends/claude/runtime.md.
  • Codex — pinned codex exec policy, loop-owned exact-id state, and policy-banner verification. backends/codex/runtime.md.
  • grok — a fail-closed custom sandbox, linked-worktree placement, and a per-machine tuple allowlist. [`backen

// HOW IT'S BUILT

KEY FILES

hosts/cursor/skills/cursor-engineering-mode/SKILL.mdREADME.md

// REPO STATS

16 stars