The Build Loop

A steady rhythm of planned, briefed, shipped slices.

Last updated 2026-08-28

Building software with Leefy is a loop, not a waterfall: a planning agent turns a branch into sliced briefs, you approve them, and then slices ship one at a time — each one briefed, run, harvested, and banked before the next begins.

Leefy stays authoritative on the what (goals, acceptance criteria, scope); the coding agent grounds the how in your actual repo at session time. The Accept gate and the premise check are where those two meet honestly.

The loop turns the same way whichever way the brief travels. By clipboard is the default: copy the slice prompt, paste it at your agent, paste the summary back. Connected takes the clipboard out of both directions — issue an agent token under More ▸ Tree ▸ Agent access, paste the setup command it gives you, and your agent reads the next slice and its brief straight out of the tree, then files its summary back itself. The human is the stop condition either way.

  1. 1

    Line up the work

    Code Mode

    Press Code Mode and let your coach read the whole tree: it groups the nodes that are ready to build into one ordered wave, names it, and says what shipping all of it achieves.

    You’re here when: Several nodes across the tree are ready to build, but nothing says which order to build them in.

  2. 2

    Plan the work

    Plan session

    Run a Plan session on the branch you want built. A coding agent explores your repo and returns an ordered set of slice briefs, each with a run-with recommendation.

    You’re here when: A branch holds the intent, but there is no ordered queue of slices for it yet.

  3. 3

    Accept the plan

    Outside Leefy

    Review the returned briefs at the Accept gate — edit what reads wrong, park what can wait, approve the rest. Accepting plants the slice queue under the branch.

    You’re here when: A plan summary has been pasted back and is waiting for your review.

  4. 4

    Connect your agent (optional, once)

    Outside Leefy

    Under More ▸ Tree ▸ Agent access, issue a token and paste the setup command for your runner — Claude Code, Grok Build or Codex. From then on the agent reads the next slice and its brief from the tree itself, so you stop ferrying prompts by clipboard.

    You’re here when: A queue exists and you are copying slice prompts out of Leefy by hand every time.

  5. 5

    Take the next slice

    Run the work

    The queue always knows what is next. Copy the slice prompt — instant, straight from your accepted brief — and hand it to the agent; a connected agent pulls the same brief itself and just starts.

    You’re here when: The queue shows an up-next slice that is not yet in motion.

  6. 6

    Ship it and mark it done

    Mark done

    Let a connected agent file its summary, or paste it back yourself, then harvest it. Watch the premise check — anything the brief got wrong shows up amber. Mark the slice done and the queue advances.

    You’re here when: A session has come back with a summary ready to bank.

  7. 7

    Review, then go again

    Review

    Skim what shipped and let it correct the plan: revise briefs that drifted, then take the next slice — or run a fresh Plan session when the queue runs dry.

    You’re here when: Slices are landing and the loop is turning; the question is only what turns it next.