- Home
- Skills
- Agents & Orchestration
- Implementator
More screenshots
Works with the AI Agents you already use
Implementator
Puts every code change through one loop in Claude Code or Codex: a plan you approve, tests, a review, your manual test, and a written record of it all.
$20
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
Implementator
Example session with this skill installed
Start work on TAL-1.
- Read your context and instructions
- Compiled the implementator
TAL-1 is open as TAL-1-add-json-flag, with a workstream called json-flag, and the plan is ready for your approval. No code has been written yet. The full plan is at workstreams/json-flag/plans/001-PLAN.md.
Summary: add a --json flag to src/cli.js that prints {"total":5,"unique":4} on one line instead of the sentence. The default output doesn't change and no dependency is added. It touches src/cli.js, the README and a new test/cli.test.js. The risk is low, and src/count.js isn't touched.
The change
const args = process.argv.slice(2);
const json = args.includes('--json');
const file = args.find((a) => a !== '--json');
if (!file) {
console.error('usage: tally [--json] <file>');
process.exit(2);
}
const { total, unique } = countWords(readFileSync(file, 'utf8'));
console.log(json ? JSON.stringify({ total, unique }) : `${total} words, ${unique} unique`);
Choices the ticket left open
--jsonworks before or after the file name.tally --jsonwith no file is a usage error: a message on stderr and exit code 2.- The usage line gains
[--json]. - Other unknown arguments behave as they do today, so the first one is treated as the file.
- The JSON has only
totalandunique, the two counts the ticket names.
Tests: test/cli.test.js is the repo's first test of the command itself, and it runs the real process. It checks the exact output with and without the flag, the flag placed after the file, an empty file, and the usage error.
Left out on purpose: the raw crash on a missing file stays as it is, with or without the flag, because the ticket doesn't ask for it. I can record it as a separate follow-up.
Do you approve this plan? After that I'll run through coding, tests, the self-test and review on my own. I'll stop again when the manual test is ready for you, and nothing gets committed until you confirm.
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
Coding agents are quick, and they skip things. They guess at your stack, commit before anyone has looked, and forget all of it by the next session. Implementator gives your agent one way of working across every repo you connect, and keeps a record of what happened in a notes repo that belongs to you.
What it does
Each change goes through the same stages: a plan, the code, your tests and checks, a self-test where the agent drives the real app, a severity-ranked review, a manual test, and the commit. Failures go back through the loop.
It scans each repo first and writes skills for that repo: how to code, test, review and commit there, with each rule tied to a file it found. Plans, reviews, test records and a dated journal stay in the notes, so the next session carries on where the last one stopped.
Where you are needed
The agent stops for you at five points: it will not write code before you approve the plan, you run the manual test from the steps it gives you, you confirm the commit, you say when to close, and you commit the notes.
Settings change how a stage runs: who commits, which browser, which model tier, draft PRs or none. They cannot switch off the plan, the review or the manual test.
What is in the pack
13 skills: setup, connecting a repo, generating its skills, preflight, the loop, closing work, settings, plus tools for questioning a plan, diagnosing a hard bug, mocking up a UI on the running app, and PR screenshots. It reads a named ticket from Jira or another tracker if your agent has that connector, and it works with GitHub pull requests and CI.
What it needs
Python 3.8 or newer and git. Optional: gh or glab for pull requests and CI, and a Playwright MCP server for browser self-tests.
Full docs, every setting and troubleshooting: https://implementator.io/
The rules are instructions, not locks: there are no local hooks, so an agent can drift, and you should read what it reports. One script does refuse to mark work resolved without a plan, a review, a self-test record and a manual run on disk. Tested so far on one small Node project, on Windows only. On Codex under Windows, Node test suites must be run by you. It uses real usage: a full run with a review by subagents costs several dollars, more on a larger change. Windows path limits apply, so keep the notes repo on a short path.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 10 days ago
Requires
Permissions
Writes only to the notes repo you choose and to the code repos you connect. Scripts run from a temporary file that is deleted afterwards. git, gh and your project's own commands run in your terminal. Pushes and pull requests go to your own remote, and a commit waits for your yes. The browser is used through your own Playwright MCP server.