Who this is for
Engineers who solve the design in their head but stall on turning it into a written spec everyone can review.
The moment this saves you
I've solved the design in my head on the walk home, then I sit down to write the spec and the blank doc kills all my momentum, so the doc never gets written and the context stays trapped in my head.
See it work
Messy spoken thought in. A clean, structured artifact out.
Okay let me talk through the spec for the notifications system. The problem is right now notifications are sent inline and synchronously, which is slow and fragile. The approach I'm thinking is an event bus, services publish events, a notification worker subscribes and handles delivery. The main tradeoff is added infra complexity, we'd need to run the bus, but we get decoupling and retries for free. Open questions, do we use the existing Redis or bring in something like Kafka, and how do we handle notification preferences, per user or per event type. I'm leaning Redis to start. Out of scope for v1 is digesting and batching notifications.
Spec: Notifications system
- Problem: Notifications are sent inline and synchronously, slow and fragile.
- Proposed approach: Event bus. Services publish events; a notification worker subscribes and handles delivery.
- Tradeoffs: Added infra complexity (running the bus) in exchange for decoupling and free retries.
- Open questions:
- Existing Redis vs introducing Kafka? (leaning Redis to start) - Notification preferences: per user or per event type?
- Out of scope (v1): Digesting and batching notifications
The workflow
Record a voice note
Hit the hotkey and talk, no formatting, no typing.
Tag it with this context
Contextli shapes your words into the structured output above.
Find it later
Everything's searchable and organised by context.
Pull it into Claude or ChatGPT
Bring your contexts straight into your AI tools with the Contextli MCP.
Your raw recording and transcription stay on your device, so you can always go back to the original.
The prompt behind this context
I'm going to talk through a technical design or spec. Turn it into a spec skeleton with: a bold Title, then labeled sections: Problem (what we're solving), Proposed approach (the design, in my framing), Tradeoffs (what we gain and what gets harder), Open questions (a bullet list of the unresolved decisions, keeping my lean if I state one), and Out of scope (only if I mention what's excluded). Keep my technical reasoning and any leanings exactly. Don't invent approaches, tradeoffs, or decisions I didn't raise. Output only the skeleton.
Make it your own. This is a starting point. Once it's in Contextli, tweak the instructions so the output comes out exactly how you like it.
Use this context
One tap adds it to your clipboard. Open Contextli and paste to add it.
Next, open Contextli, Contexts, Import, paste.
Make it your own. This is a starting point. Once it's in Contextli, tweak the instructions so the output comes out exactly how you like it.
Your raw recording and transcription stay on your device, so you can always go back to the original.
Related contexts
Code Decision Log
Right after you make the call, while the tradeoffs are still fresh, talk through why you went this way and what you rejected. Three months from now when someone asks 'why a queue here?', the answer is already written down.
Feature Request
A customer asks for something and it lands in your head as a vague 'they want X.' Say what they actually asked and the problem underneath. You get a brief with the real user problem, not just the feature, so the backlog reflects needs instead of demands.
On-Call Incident Note
You just fought a fire at 2am and the postmortem is tomorrow, when half the timeline will be gone. Talk through what broke, what you did, and what fixed it, right now. You get a postmortem-ready note while the details are still sharp and real.