Skip to content
tuckit

The daily loop

Attaching an agent doesn’t make the board true. A habit does. There are three parts, in order.

Starting: make it read the board

Section titled “Starting: make it read the board”

Anything shaped like where are we, what should I do next, or what did we decide about this is answered by the board, not by the code.

Commit history tells you what changed. It says nothing about what was decided and not yet built. Ask an AI “what’s next?” and it will usually go digging through code anyway, answer from information that isn’t there, and sound completely sure of itself.

Ask like this:

What's the state of this project?

With the plugin installed, a note goes in at the start of every session telling it to do exactly this, so most days you won’t have to. When you do:

Read the current state from tuckit before you answer.

Before starting work: check for an existing card

Section titled “Before starting work: check for an existing card”

Before doing anything real, check whether a card already covers it. It might be filed in an area, or sitting in the Inbox.

  • If one does, continue that one. Read its Constraints first. That’s where the last person marked the landmines.
  • If none does, make it before you start, not after.

This is nearly always why boards go stale: work the board doesn’t know about. Half-finished work that lives only in your terminal is invisible to your teammates, to you next week, and to the next session.

The Inbox deserves a specific look. The count is always visible, but unsorted items are the easiest to miss and usually the oldest.

Show me what's sitting in the Inbox.

The column tells you what to ask for

Section titled “The column tells you what to ask for”

You don’t have to open a card to know what it needs. Every card reports which column it’s in (Stage), and each column names one action.

Column What to do next
needs_design Work out what to build, and write it into the card
needs_steps Break it into an ordered list of steps
executing Do the next unfinished step and tick it off
ready_to_ship Nothing automatic. A person calls it finished

This is also the honest way to ask for a plan. “What should we work on next?” invites invention. “Which cards haven’t been worked out yet?” has an answer.

The Hand to an agent button on a card writes the right sentence for whichever column it’s in. Copy, paste, done.

Finishing: make it write

Section titled “Finishing: make it write”

This is where boards break. All of it is cheap, and skipping it turns a record into a lie.

What comes next matters most. If the session ended on “we’ll do the frontend part after this”, that’s a card. Make it now, in the right area, with the “what we’re building” box left empty so it correctly reads as not yet worked out. A vaguer someday idea is a card with no area, which lands in the Inbox. An agreement that isn’t on the board isn’t an agreement.

Bugs and follow-ups noticed along the way just get written down. Don’t try to file them perfectly; leave them in the Inbox. Writing things down without sorting them is a proper path precisely so noticing stays cheap.

What actually happened goes in two places: tick off the steps you finished, and leave a note on the card. Notes are past tense: what you did, what blocked you, where it landed. Ticking the last step already moves the card to ready_to_ship, so there’s nothing else to set.

Calling it finished is yours. That’s a judgement, not a calculation, so an agent should propose it and you should make it.

With the plugin installed this arrives as a nudge at the end of a session. You can also ask for it at any point:

Reconcile the board with what we just did

With another AI, say it plainly:

Write what we just did back to the tuckit board, and create cards for
anything we said we'd do next.

Pointing at one specific card

Section titled “Pointing at one specific card”

Every card has a number like ACME-42, copyable from anywhere in the app. Paste it straight into a chat:

Pick up ACME-42.

Working alongside other people

Section titled “Working alongside other people”

Several people and several agents can hold the same board at once. Two habits keep that pleasant.

Leave notes on cards instead of reporting in chat. A note is visible to everyone who opens the card later. A message in your terminal isn’t.

Use external_key for repeating automation. If a script or a scheduled agent creates cards, giving each a stable key makes a re-run update the existing card instead of making a second copy. The details are in MCP tools.

Where to go

Section titled “Where to go”