Cook with Pan Out
This is the part where the screen gets splattered. You have a protocol, your ingredients are on the counter, and you’re ready to go.
cook beef stew
Before the heat goes on
The cook skill doesn’t throw you straight into chopping. It runs through a startup sequence:
Scale and substitutions
First question: “How much are we working with?”
Answer however makes sense — “I’ve got 1.2kg of chuck”, “feeding 6 tonight”, or “just a small batch.” The skill takes whatever you give it and scales the protocol’s quantities.
Then: “Any ingredients you’re missing or want to swap?” If the protocol lists substitutes (e.g., shallots for onions), it’ll mention those. If you want to swap something not on the list, it reasons about the impact and adjusts.
Audio check
The skill tests whether voice works on your system. You’ll hear “Can you hear me?” — confirm, and voice prompts are on for the whole session. If TTS isn’t working, it falls back to chime alerts, then screen-only as a last resort.
Voice is important. When your hands are covered in raw meat and you’re managing a hot pan, a two-sentence spoken summary beats scrolling through a chat window.
Pre-flight briefing
Before each phase starts, the skill runs an aviation-style checklist:
- What this phase is about and why it matters
- Equipment you’ll need ready
- Questions? “Now’s the time.”
Two modes of cooking
The cook skill has two modes, and it switches between them automatically based on what the protocol says about each phase.
Pull mode — you’re at the stove
Active phases (prep, sear, saute) run in pull mode. The skill gives you one instruction at a time and waits for you to confirm before moving on.
Voice: “Sear the first batch — 10 to 12 cubes, 90 seconds per face.”
Screen shows the full detail: batch size, sensory cue (“mahogany brown crust, releases from pan without sticking”), the science behind why crowding kills the sear.
You do the step, say “done” or “next.”
The skill advances. “Step 3 of 5. Remove the beef, saute the onions in the fond.”
If you ask a question mid-step — “what’s an oblique cut?” — the skill answers immediately, with a full mechanical how-to, then picks up where you left off. Questions always take priority over advancing the protocol.
Push mode — you can walk away
Passive phases (braise, rest, marinate) run in push mode. A timer starts in the background, and the skill tells you what to expect:
Voice: “Braise is on. Timer set for 90 minutes. You can walk away — I’ll call you back for a lid-lift check.”
During the hold, the skill does useful things with the dead time:
- Pre-flight briefing for the next phase — what to prepare, what to have ready
- Science context — why collagen converts at 80-90C, why a lazy simmer beats a rolling boil
- Sensor polls at regular intervals — “What’s the thermocouple reading?”
When the timer fires, it calls you back:
Voice: “Timer’s up. Lid-lift check — grab your thermometer.”
It asks for a sensor reading, evaluates whether to continue the hold or transition, and moves on when the food is ready — guided by sensory cues, not just the clock.
Talking to it
You’re cooking, not typing an essay. The skill understands natural confirmations:
- “done”, “next”, “ready” — advance to the next step
- “what’s an oblique cut?” — get a technique explainer
- “timer went off” — skip to timer completion handling
- “I messed up” or “it’s burning” — triggers error recovery
- “what’s next?” — preview what’s coming
If something goes wrong, the skill stays calm. It assesses impact (“That’s maybe a 3 out of 10 on the final dish”), gives you a forward path, and logs the deviation so the debrief skill can learn from it.
Temperatures
If you set up sensor calibration, temperature guidance looks like this:
“We want 90C actual — that’s about 86-87C on your thermocouple.”
Both values, every time. The protocol stores true temperatures; the skill reads your calibration data and does the math so you never have to. Without calibration, you just see the true target.
Session state
Every cook is tracked in a session state file (sessions/cook-YYYY-MM-DD-dish.md). This is automatic — you won’t notice it happening. But it means:
- Crash recovery — if Claude Code restarts mid-cook, it can pick up where you left off
- Debrief fuel — the state file captures timestamps, sensor readings, deviations, and decisions, giving the debrief skill rich data to work with
Wrapping up
When the last phase completes, the skill offers serving guidance, storage tips, and reheating instructions. Then:
Voice: “Nice work. That’s a wrap.”
It offers to run a debrief while the cook is fresh in your mind.