add
ChatGPTAdd a task, a request for the human, or a question for the human. It stays open: claim it only if you start on it now, and tick it once the work itself is done, not for adding it.
answer
ChatGPTRecord the answer the human gave you in chat. Never pick one yourself (use assume) or answer to unblock work (use edit's needs).
assume
ChatGPTCan't wait for the human? Pick an option and go on; the question stays open, marked assumed, until the human confirms or overturns it.
board
ChatGPTShow the human this project's board in the chat: the questions and requests waiting on them, with each question's options, who is at work on what, and what waits to be looked over. They can answer and tick from it, and what they do there is done. Use it when they ask what's going on, what waits on them, or to see the board.
claim
ChatGPTSay you're working on an item, so other agents pick something else. It lasts until you've been quiet for 30 minutes; tick drops it when done, release: true if you stop without finishing it, after a note on what you did and ruled out (edit's add_note). Items for someone else are refused.
edit
ChatGPTChange an item's list, title, section, fields, needs (it's blocked until they're done), notes, a question's options, or its place in the list. delete only when the human asks, or calls the work off, or for a duplicate you made; restore brings it back.
import
ChatGPTBring work kept somewhere else (a TODO or plan file, notes, issues, a checklist, your own todo list, another tool's lists) into the lists in one call, once the human says yes. Sort it as it comes in: what only the human can do (an account, a key, money) is a request, a decision of theirs a question, each in that list and not also a task. Write each list's items as Markdown, done ones too: Heading, - [ ] Title (- [x] when done), indented lines under it its notes, indented - [ ] its subtasks. A question's options are its indented - lines, (recommended) after one; its notes there have no dash. A file's title is left out; text before the first item is the list's intro. Notes hold up to 20,000 characters an item. What else the source says goes in a comment ending the line, all of it: <!-- id:T1 by:claude at:2026-10-01T09:00 needs:T2 prio:high done:2026-10-02 done-by:you answer:"Pick-up only" answered-by:you -->; needs names items by those ids or by keys here. Lists already in this form (another tool's export) go in as they are. A log of past work (test rounds, a changelog) stays in its file: bring in only the open work it names. Work found in two places is one item; group it by what it is about, not by where you found it. An item its list already has by title is left out. All of it comes in, or nothing, into this one project: another repo's work goes into that repo's own project.
list
ChatGPTThe project's open tasks, requests and questions; key reads one item, open or done. Answer every question about the work and the lists with it. Shows 50 items at a time and says where the next ones start.
lists
ChatGPTAdd, rename or remove one of the human's own lists (ideas, bugs) when they ask, rather than a heading in the backlog; list shows them all. Its items work like tasks; pass its name as kind.
logo
ChatGPTSet the project's logo from an image in its repo, only once the human says yes to that file: find the project's own mark at its largest (an app icon or brand image before a small favicon) and show it to them first. Offer it once when the project has none; a no (no: true) stops only the offer. The reply gives an address to send the file to.
message
ChatGPTTell another agent something: you need its item, you changed something it builds on, a finding. It reaches them in their news under their next tool reply; no claim is needed to send one.
new_project
ChatGPTStart a new project of your own, with empty lists, for work that belongs to none of your projects yet, such as a new repo. Name it after that work. The other tools then take it as project.
news
ChatGPTWhat changed since you last looked: the human's answers and edits, other agents' claims, ticks and notes, and messages to you. Each is told once.
projects
ChatGPTList the projects you are on, with what is open for the human in each: the place to start for what waits on them. Each has its own lists, and every other tool takes one of them as project.
section
ChatGPTRename, reorder or remove a heading in a list (add makes a new one: give an item that section). Give one of to, before, after or remove.
start_here
ChatGPTYour human's answer to whether the board may start Claude Code or Codex for this project on this computer, in this folder, when they give a task to a new assistant. It is off until they say yes. Never on your own: ask them, and call this with yes: true once they say yes, or no: true if they say no; a no is kept so no assistant asks again here, and they can still say yes later. Only Claude Code or Codex on a computer linked to Hamloy can call it. They turn it off again on their account page.
tick
ChatGPTMark a task or request done (done: false opens it again). Drops your claim on it.
view
ChatGPTMake, change or remove a saved view of a list on the page when the human asks to see the lists a certain way, and say what it shows; no arguments lists them. E.g. most urgent first: layout table, sort [-prio]; a board: layout board, group prio, order [high, medium, low]; blocked: filter [blocked=waiting]; everything for Sara: list all, filter [for=Sara]; the messages between agents: list history, filter [what=message]; a column for each agent: list history, layout board, group who. Fields: the items' own or section, done, blocked, claimed, by, at, doneat, needs, list; of the history: what, who, to, item, list, file, branch, at, text, read (whether a message reached its agent). Fix what the reply warns about.