Daily task guide.
For working through your P1 / P2 task list. Open this when you're about to do work.
Quick reference
- Open ECB Company Work · paste your prompt · get briefing
- Tell Claude which task you want · it tells you which model + gives you a task brief (only for complex tasks)
- Take brief to LYRA · paste · copy the optimised version
- Back to your briefing chat · switch model at the bottom · paste optimised prompt · run
- Post the Slack handoff message in #p0-tasks when done
The flow at a glance
From "what should I work on?" to finished work.
Step 5 is optional — only complex tasks need the Lyra round-trip. Claude tells you when. Simple tasks go from Step 4 straight to Step 6.
Step 1 — Find ECB Company Work
Go to claude.ai and sign in with the shared ECB Google account — info@epiccomedyberlin.com. We all use this same login for Claude (and most other ECB platforms). If you don't have the password, ask Aaron.
Once you're in, look at the left sidebar. Under Starred, click ECB Company Work:
Step 2 — Your prompt
You have a personal prompt. Save it on your phone, in your notes app, wherever — copy and paste it whenever you want a briefing.
Always start a fresh chat. Every time you want a briefing, click ECB Company Work and start a new conversation. Never continue an old briefing chat for a new day's work.
Why: Claude loads the latest task files at the start of each new chat. Old chats are stuck with whatever was loaded the day they began.
Step 3 — Read your briefing
Claude returns your tasks split by priority. You'll see P1 and P2 tags on each task.
Each task has a one-line "next concrete action" — the smallest step that moves it forward today. Pick one and tell Claude you want to work on it.
Step 4 — Pick a task
Tell Claude which task you want to work on (e.g. "Let's do the newsletter task" or "I'm ready to start #11"). Claude assesses how complex the task is and either starts working with you directly, or hands you a task brief to take to Lyra first.
For simple tasks — Claude just starts
Quick admin, scheduling, lookups, basic emails — Claude tells you to stay on Sonnet 4.6 and works through it with you in the same chat. Skip ahead to Step 6.
For complex tasks — Claude gives you three things
Anything creative, strategic, copy-heavy, research-driven, or high-stakes (newsletter, contract review, ad copy, competitor work, funding application) — Claude outputs (1) a model recommendation, (2) a task brief to take to Lyra, and (3) instructions on what to do next. The optimised prompt comes back to this same chat — you don't open a new one.
This second route exists because the briefing chat is just running off your task list. Lyra takes that rough task and turns it into a precision-crafted prompt that gets you a meaningfully better result.
Step 5 — Optimise via Lyra (complex tasks only)
Lyra is a prompt specialist — a saved chat in your sidebar called LYRA that turns rough requests into precision-crafted prompts. You'll round-trip through it any time Claude hands you a task brief.
What Claude gives you in the briefing chat
When you pick a complex task, Claude in the briefing chat outputs three things together: a model recommendation, a task brief in a code block, and instructions on what to do next. Here's what the task brief part looks like:
That whole block is your task brief. Claude generates it for you — you don't need to write it.
The round-trip — step by step
Copy the task brief
Select the whole code block (everything inside the dark box). Copy it.
Open the LYRA chat
Find LYRA in the Starred section of the sidebar. Click it.
Paste and send
Paste the brief into the chat box. Hit enter. Lyra usually gives you back an optimised prompt straight away. Occasionally — if it's a DETAIL request and it genuinely needs more context to do a good job — it'll ask one or two clarifying questions first. Answer them, and you'll get the optimised prompt next. The first word in the brief — BASIC or DETAIL — tells Lyra how thorough to be.
Copy the optimised prompt
Lyra outputs a block called "Your Optimized Prompt:" — copy that whole block.
Go back to your briefing chat
Click into the sidebar. Under Recents at the top of the chat list, you'll see your briefing chat — it's the one you were just in. Click it. Don't open a new chat — go back to the same one.
Switch the model before pasting
This is the part most people miss. Look at the bottom of the chat input box — you'll see the current model name (e.g. "Claude Sonnet 4.6") with a small dropdown arrow next to it. Click that name. A list of models opens. Pick the one Claude told you to use (e.g. Opus 4.7 for the newsletter example above). The change takes effect from your next message onwards.
Paste the optimised prompt and send
Paste the prompt Lyra gave you into the chat box. Send it. Claude — now running on the right model — will start working through the task with you in guided mode.
BASIC — quick optimisation. Lyra returns a polished prompt immediately. Used for routine creative work.
DETAIL — for high-stakes work where the framing matters. Lyra builds a more comprehensive, carefully-scoped prompt. It will ask clarifying questions only if it needs missing context — most of the time it has enough to work with and just returns the optimised prompt straight away.
Claude in the briefing chat decides which to use when generating your task brief. You don't need to think about it — just paste what you're given.
Step 6 — Work with Claude
From here on, everything happens in your briefing chat — same chat the whole way through. Whether you went through Lyra or skipped it, Claude is now in guided mode. It asks you one thing at a time, pulls research from credible sources (named experts, primary studies, recognised industry authorities), and walks you through the work step by step.
You don't need to know what comes next. Just answer what Claude asks. Your job is to bring the voice, the instincts, the decisions — Claude handles the structure, drafts, and frameworks. When Claude cites research, it'll name the source so you can judge for yourself.
You're working on the newsletter. Claude reads about newsletter best practices, names the experts it's drawing from, asks what voice you want, presents a structure, you rewrite a sentence in your voice, Claude folds your version in, asks about subject lines, you pick one, Claude drafts the first issue. By the end you have a finished newsletter without ever needing to plan the steps yourself.
For anything that matters — a draft going out, a decision recommendation, a final deliverable — explicitly ask Claude to verify before you accept it. Claude often catches its own errors when asked to look. It just doesn't always do it on its own.
Use lines like: "Check your work — list every requirement I gave you and confirm each one is covered" · "Verify the facts in this draft against the project files" · "Is anything in this output unsupported or guessed?"
Two minutes of self-check saves an hour of fixing it later.
If you need to pause
You don't have to finish a task in one sitting. Real work has interruptions — meetings, blockers, needing input from someone else, just running out of time. You can always stop and come back.
To stop for now
Just close the tab or walk away. Claude saves the chat automatically. Nothing's lost.
To resume the same task later
Go back to claude.ai. On the left side of the screen you'll see a sidebar — if it's hidden, click the small menu icon at the top-left to open it. Scroll down past Starred and you'll find a section called Recents. Your chat will be near the top (it's the most recent thing you worked on). Click the chat title to open it. Everything is still there — your conversation, the work in progress, all of it. Just keep going from where you left off.
To start a different task
Click ECB Company Work in the Starred section of the sidebar (same as before), and paste your prompt. That gives you a fresh briefing. Different chat, new context, doesn't affect the paused one.
Resume same task → sidebar → Recents → click your chat
Start a new task → sidebar → Starred → ECB Company Work → paste prompt
Step 7 — When you finish
When the task is done, Claude tells you. It also gives you a short message to post in #p0-tasks on Slack. Post it. That's the whole handoff.
Aaron sees the post, reads your chat, and updates the task files. Your work shows as complete in the next briefing. You don't need to do anything else.
Got extra time?
Your briefing leads with P1 and P2 because those are what's most pressing. But every task in the system has a priority — and when your urgent work is clear, there's still useful work in the backlog.
If you want to see them, ask Claude: "What's on my P3 list?" or "Show me my full task list." Claude will pull them and walk you through any one you pick — same process as P1 and P2, including the Lyra round-trip when needed.
The models
Claude comes in three sizes. Think of them as different versions of the same brain — some are faster, some go deeper. Default is Sonnet 4.6. Claude tells you when to switch as part of any task it gives you.
For almost everything. Briefings, writing, drafts, social, email, research. Start here.
For deep work — contract review, complex strategy, long document analysis, anything that needs sustained reasoning.
For quick lookups or simple questions. Rarely the right pick for ECB work.
How to switch
Look at the bottom of the chat input box (where you type). You'll see the current model name (e.g. "Claude Sonnet 4.6") with a small dropdown arrow next to it. Click the model name, pick the model you want from the dropdown, then send your next message. The change applies from your next message onwards — your conversation history is preserved.
Opus costs ~5× more per token than Haiku, and ~1.5–2× more than Sonnet — and Opus also "thinks" longer, so a long Opus chat can chew through 10× what the same chat on Sonnet would. We share one Max plan, so wrong-model use locks the whole team out until the 5-hour reset.
Default Sonnet 4.6. Only switch to Opus when Claude tells you to or when you're starting genuinely deep thinking up front. Drop back to Sonnet once the heavy reasoning phase is over.
Full breakdown — including the "plan with Opus, execute with Sonnet" pattern for saving tokens mid-chat — is in the General Guide.
Trust Claude's recommendation in your task briefing. If Claude doesn't say to switch, stay on Sonnet 4.6 — it handles 90% of ECB work fine.
Don't switch models routinely mid-chat — switching costs tokens (the new model re-reads the whole chat history before responding). The only time it's worth it: you started on Opus for genuine deep thinking, that phase is clearly over, and you have significant execution work ahead. Then drop to Sonnet, once. Otherwise, stay where you are. Full breakdown in the General Guide.
Coordinating with the team
Because we share one Max plan account, your model choices and chat habits affect everyone. Two simple rules:
#p0-tasks is for tasks only — task completion, status updates, task-specific handoffs. Nothing else.
Any other team communication goes in #p1-important — heads-up posts, dependency flags, coordination, decisions, things people need to know. Don't post these in #p0-tasks.
Heavy task coming up — flag it
If your task is time-sensitive AND will use a lot of Opus, post in #p1-important before you start so nobody else fires a non-urgent Opus session at the same time:
Naming dependencies after meetings
If your task blocks someone else's, say so explicitly when you start — don't assume people remember from the meeting. Dependency reminders go in #p1-important, not #p0-tasks:
Full guidance on shared-account coordination is in the General Guide.