Accoil Logo
View on GitHub
Product Tracking Skills

Your B2B SaaS isn't data-ready

Open-source AI agent skills that scan your codebase, design an opinionated tracking plan, and generate working instrumentation code — for any analytics SDK.

View on GitHub

or install directly

Works in

Most B2B products are flying blind

You're paying for Amplitude, Mixpanel, or PostHog. But you can't answer basic questions:

  • Which features do your customers actually use?
  • Where do users drop off in your core workflow?
  • Which accounts are at risk of churning?

The problem isn't your analytics tool. It's your instrumentation. Your tracking is missing, inconsistent, or decayed into chaos — and no one has time to fix it.

The cost of staying blind

Product decisions

Made on gut feel, not data

Customer success

No visibility into account health

Investor conversations

No credible usage metrics

Engineering time

Wasted on ad-hoc tracking requests

Analytics ROI

Zero — because the data going in is garbage

Six skills. One session. Data-ready.

Product Tracking Skills walks you through a structured 6-phase process — from understanding your product to generating real, typed, SDK-specific code.

Each skill produces artifacts that feed the next. Everything lives in your repo. Context survives across sessions and engineers.

1

Model (what)

Scans your codebase, asks targeted questions about value and entities

2

Audit (is)

Reverse-engineers current tracking from your code

3

Design (should)

Designs an opinionated target plan + explicit delta

4

Instrument (how)

Maps the plan to your specific SDK

5

Implement (apply)

Generates real code following the guide

6

Maintain (evolve)

Updates tracking when features ship

1

Model

Scans your codebase, asks targeted questions about value and entities

Product model: what your product does, who uses it, how value flows

2

Audit

Reverse-engineers current tracking from your code

Factual inventory of what's actually tracked today

3

Design

Designs an opinionated target plan + explicit delta

Tracking plan + the exact diff from current to target

4

Instrument

Maps the plan to your specific SDK

SDK-specific instrumentation guide with template code

5

Implement

Generates real code following the guide

TypeScript types, typed wrapper functions, usage examples

6

Maintain

Updates tracking when features ship

Versioned plan updates + changelog

Three commands to data-ready

1

Install the skills

Add the skills to your AI agent tool. Works with Claude Code, Codex, VS Code — any tool with agent support.

2

Point it at your codebase

Say "model this product" and the skills scan your routes, models, and controllers to understand what your product does.

3

Run the lifecycle

Audit what's tracked. Design what should be tracked. Generate the code to close the gap. Done.

terminal
You: model this product
AI: [Scans codebase, asks about value and entities]
Product model saved to .telemetry/product.md
You: audit tracking
AI: Found 14 events across 8 files.
Saved to .telemetry/current-state.yaml
You: design tracking plan
AI: Target: 22 events. Delta: add 10, rename 3, change 4, remove 1.
You: implement tracking
AI: Instrumentation code ready in instrumentation/

Built for the people who feel the pain

Founders

You know you need product analytics. You've been meaning to set it up for months. You've probably tried once and ended up with five inconsistently-named events. These skills do it properly in a fraction of the time.

Product Engineers

Tracking code scattered across 23 files. Mixed casing. No one knows what's tracked. You need a system: audit reality, design intent, generate typed code, maintain as features ship. That's what this is.

CS / Success Teams

You need visibility into how accounts use the product. Feature adoption. Engagement signals. Churn risk. You can't build it yourself, but you can hand engineering a clear process: "Run these skills. Get us the data."

Users and accounts are first-class

This isn't web analytics adapted for B2B. It's built for products where users belong to accounts, accounts have hierarchies, and feature adoption matters at the account level.

  • Group hierarchy built in (account > workspace > project)
  • User and account traits designed as part of the plan
  • Event attribution at the correct group level
  • Identity management (identify, group, reset) handled per SDK

Supports six analytics SDKs (so far)

Segment logo
Segment
Amplitude logo
Amplitude
Mixpanel logo
Mixpanel
PostHog logo
PostHog
Accoil logo
Accoil
RudderStack logo
RudderStack

Not a prompt. Not a template. A process.

Ask an AI chatbot about tracking
Generic advice. No codebase awareness. No structure. Forgotten by next session.
Tracking plan spreadsheet
A document that drifts from reality on day one. No code generation.
Auto-capture (Heap, FullStory)
Clicks and pageviews. No semantic meaning. No B2B account context.
Hire an analytics engineer
Expensive. Slow. Knowledge walks out the door.
Product Tracking Skills
Scans your codebase. Designs an opinionated plan. Generates real code. Maintains itself. Lives in your repo.

Everything lives in .telemetry/

Every artifact is a file in your repo. Git tracks how your understanding evolved. Nothing lives in someone's head or a third-party tool.

file tree
.telemetry/
├── product.md              # What your product does
├── current-state.yaml      # What's tracked today
├── tracking-plan.yaml      # What should be tracked
├── delta.md                # Current → target diff
├── instrument.md           # SDK-specific guide
├── changelog.md            # How the plan evolved
└── audits/
    └── 2026-02-13.md       # Audit snapshots

We take positions so you don't have to debate them

  • dot.notation for events. report.created, not Created Report or reportCreated. Object then action, always.
  • snake_case for properties and traits. plan_type, company_size, last_seen_at. Consistent, predictable, queryable.
  • Properties over events. One event with a type property beats three similar events.
  • Minimalist coverage. Track what matters. Skip what doesn't. Less is cheaper.
  • Audit before design. Describe reality before deciding intent. No assumptions.
  • Delta-driven. Current → target diff is the implementation backlog. No ambiguity.

Opinionated starting points for common product types

B2B SaaS Core

Generic B2B SaaS baseline

AI/ML Tools

AI products, generation, models

Developer Tools

APIs, SDKs, CLI tools

Collaboration Tools

Team workspaces, real-time collab

Form Builders

Form creation, submissions

Security Products

Security events, alerts, compliance

Analytics Platforms

Analytics products tracking their own usage

Frequently asked questions

Your product has things worth tracking. Start now.

MIT-licensed. Install in 2 minutes. Run "model this product" to begin.