CtrlK
BlogDocsLog inGet started
Tessl Logo

mcollina/init

Creates, updates, or optimizes an AGENTS.md file for a repository with minimal, high-signal instructions covering non-discoverable coding conventions, tooling quirks, workflow preferences, and project-specific rules that agents cannot infer from reading the codebase. Use when setting up agent instructions or Claude configuration for a new repository, when an existing AGENTS.md is too long, generic, or stale, when agents repeatedly make avoidable mistakes, or when repository workflows have changed and the agent configuration needs pruning. Applies a discoverability filter—omitting anything Claude can learn from README, code, config, or directory structure—and a quality gate to verify each line remains accurate and operationally significant.

85

1.14x
Quality

94%

Does it follow best practices?

Impact

72%

1.14x

Average score across 5 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

task.mdevals/scenario-2/

Task

We're a platform engineering team of 40 engineers working in a monorepo. The repo has grown significantly over the past two years and now has 8 packages covering everything from frontend to ML inference to payment processing. We want to set up AI agent configuration so that assistants can help effectively across the whole codebase—knowing what commands to run, what gotchas exist, and how to work safely in each part of the repo.

Please set up the agent configuration for us. Below is the relevant repo context.


Repo context

package.json (root)

{
  "name": "acme-monorepo",
  "private": true,
  "workspaces": [
    "packages/api",
    "packages/web",
    "packages/payments",
    "packages/ml",
    "packages/shared",
    "packages/admin",
    "packages/notifications",
    "packages/analytics"
  ],
  "scripts": {
    "build": "turbo run build",
    "test": "turbo run test",
    "lint": "turbo run lint",
    "dev": "turbo run dev --parallel"
  },
  "devDependencies": {
    "turbo": "^2.0.0",
    "typescript": "^5.4.0"
  }
}

README.md (root)

# Acme Monorepo

This repository contains all services and packages for the Acme platform.

## Packages

- **api** — Core REST API service
- **web** — Customer-facing React frontend
- **payments** — Payment processing service
- **ml** — Machine learning inference service
- **shared** — Shared utilities and types
- **admin** — Internal admin dashboard
- **notifications** — Email/SMS notification service
- **analytics** — Analytics event pipeline

## Getting started

Install dependencies: `npm install`
Run all services: `npm run dev`
Run tests: `npm test`

packages/api/README.md

# API Package

The core REST API built with Express and TypeScript. Exposes endpoints consumed
by the web frontend and external partners.

## Development

Start the dev server: `npm run dev` (runs on port 3001)
Run tests: `npm test`
Run linter: `npm run lint`

## Environment variables

Copy `.env.example` to `.env` and fill in values. All required variables are
documented in `.env.example`.

## Database migrations

Run pending migrations: `npm run migrate`
Roll back last migration: `npm run migrate:rollback`

packages/payments/README.md

# Payments Package

This module handles all credit card processing and billing operations for the
Acme platform using Stripe.

## PCI-DSS compliance

**This package is in PCI-DSS scope.** All changes to this package must be
reviewed by the security team before merging. Open a PR and add
`@acme/security` as a required reviewer—do not merge without their approval.

## Running tests

Tests in this package must be run against the PCI sandbox environment. The
flag `--pci-sandbox` must be passed to the test runner for all local test
runs:

npm test -- --pci-sandbox

Note: CI enforces this for the deployment pipeline, but **local test runs do
not automatically apply this flag**. Omitting it will run tests against
real Stripe test keys, which is not permitted under our PCI controls.

## Stripe configuration

Set `STRIPE_SECRET_KEY` and `STRIPE_WEBHOOK_SECRET` in your `.env` file.
Use sandbox keys for local development (prefixed with `sk_test_`).

packages/ml/README.md

# ML Package

This package provides PyTorch-based model inference for Acme's recommendation
and ranking features.

## GPU memory management

Before running any training scripts locally, you must set the following
environment variable to prevent CUDA out-of-memory errors:

export PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:128

If this is not set, training runs will frequently OOM and crash mid-epoch,
especially on machines with less than 24 GB of VRAM. This is not enforced
by any config file—you must set it manually in your shell or `.env`.

## Model weights

Trained model weights are **not stored in this repository**. They live on the
shared dev server at `/mnt/ml-storage`. You will need access to the dev
server to run inference locally. Contact the ML team to be added to the
access list.

## Running inference

python -m ml.serve --model-path /mnt/ml-storage/models/latest

## Running training

export PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:128 python -m ml.train --config configs/default.yaml


Output

Please produce the AGENTS.md file or files needed to configure AI agents working in this codebase.

evals

SKILL.md

tile.json