Skip to content

Objectives and the plan

Cora is each mission’s mission coordinator, an AI that runs the work (D0851). Every mission starts from its objectives: what it should achieve, and by when. Any member of your organisation can start a mission by giving it a name, its objectives and the people involved. Cora then starts work straight away. On her first run she writes up the full objectives as a draft, makes a first plan of work as a to-do list, and asks the mission’s lead to check the draft. Everyone you tagged in can see the mission. Nobody else can.

The examples on this page follow an invented company, Kestrel Cycles, which sells e-bikes. Its team is getting a new city e-bike into the shops for spring.

A mission (formerly project) is a chat with a job. Each mission has its objectives, its own spark tile icon, its own people, its own files and its own memory.

It helps to think of your organisation as an agency. This is an analogy, not a feature, and nothing in the console is called “agency”. In that picture:

  • Cora dispatches. She reads what is asked, answers or does the work, and calls in help when a job needs it.
  • Each mission has a job. Launch the new city e-bike, tidy the customer list, plan the next quarter.
  • Some missions become heroes. A mission’s admins can promote it to a hero: a specialist the whole organisation can call in by its own name. Another mission asks it for something, and the hero does the work with its own skills, tools, connections and keys, then sends back its answer (a summary by default, or a full answer if its admins choose). The hero’s files and documents stay in its own mission; the caller gets only the answer.

See the glossary and the hero world.

Any member of the organisation can create a mission. You do not need to be an org admin. Press New mission above the missions list, or + Create a mission on Home. You give three things:

  • A mission name, such as “Spring city e-bike”. You can rename it later in mission details.
  • The mission objectives: what the mission should achieve, and by when. The form asks “What should the mission achieve, and by when?” For example: “Launch the new city e-bike in April, with a new page on our website and an email to past buyers, and book 300 test rides by the end of May.”
  • The people tagged in: the members of your organisation who need to be part of it.

The New mission page says what happens next: “Cora, your mission coordinator, starts as soon as you create it.” She writes up your objectives in full and makes a first to-do list. You are the mission’s admin, and can make others admins later. A mission is not a hero: if it proves itself, its admins can make it one later.

When you press Create mission, the platform does everything in one go. It creates the mission, adds you and the people you tagged in as its members, creates its chat, and tells Cora to start. Either all of that happens, or nothing is created. If Cora cannot be told straight away, the start signal is kept and sent again shortly, so it is never lost.

The mission uses your organisation’s working language. You can only tag in people who are already in your organisation. If you tag someone who is not, nothing is created and the console says why.

Cora works from what you write, so a few clear lines help her most.

  • Say what done looks like. “Book 300 test rides” is clearer than “get more interest”.
  • Give a date. “By the end of May” tells Cora what to plan around.
  • List each outcome if there is more than one. Cora lists each one in the full objectives.

You do not need to cover everything. Where Cora would otherwise be guessing, she leaves a gap marked as an open question in the full objectives, instead of making something up.

The objectives show in the Overview tab of mission details (press Details at the top right of the mission). A mission’s admins, and an org admin who is in the mission, can press Edit beside Objectives, change the text and press Save (D0852). An empty box is refused with “Write the objectives in a few words.” The first of the four steps in Overview reads “Objectives written on” and the day the mission was made.

Everyone you tagged in, and you, become members of the mission. Later, a mission’s admins can add other people from the organisation: press People in the mission’s header, then Add people (D0623).

A mission is private to its members. To anyone else in the organisation, it does not exist: if they try to open it, they get “not found”.

People are also brought into a mission through their roles. When Cora needs someone in a role, the platform finds the right person and, if they are not yet a member, adds them. See roles for how that works.

At Kestrel Cycles, Ella creates the spring mission and tags in Tom, who runs the shop floor, and Grace, who looks after the website. Later, Ella adds Ben, because the new web page needs words.

Every mission has one dialogue, its chat. It is the mission’s official record of what was asked, answered, decided and done. It is created with the mission, at the same moment, and everyone tagged in is in it from the start. Anyone added later joins it too.

Everyone in the mission writes in the dialogue in their own name, and so does Cora. You can send a message as you typed it, or ask Cora to polish it first: its wording goes into your box, and you send it only when you are happy with it. Posts are never edited, so the record stays whole (D0545).

See the mission chat for how it works, and polishing a message with Cora for how Cora tidies a message before you post it.

Creating the mission starts Cora’s first run. Cora handles a mission’s work one item at a time, in order.

  1. Cora picks up the mission. It loads core-pm, the general project management skill pack. The pack’s version is pinned to the mission when it is created, so a later update to the pack never changes a mission that is already running.
  2. Cora writes up the full objectives. From the mission’s name and objectives, it writes brief.md in the mission’s documents. The full objectives stay a draft until a holder of an admin role agrees it (the fallback decision role if nobody else is set). Cora can never mark it agreed itself.
  3. Cora makes a to-do list. This is her first plan: the next five to ten pieces of work, each with the role that owns it, a rough size and what done looks like. It is part of Cora’s working memory, not a mission document.
  4. Cora asks the mission’s lead to check the full objectives. It looks up who in the mission holds the Project Lead role (or a role that runs the work) and asks that role, by the role’s own name in your catalogue. If nobody in the mission holds one, it asks the mission’s admins. It never asks a role your organisation does not have, so the question does not end up with the fallback decision role by mistake (BG-01M7K34S).

When the answer comes back, Cora writes a new version of the full objectives that records it, with the person’s name and role, and posts a short update in the chat to say what she did with it.

Cora saves its progress after each burst of work. If the machine running it stops part way through, Cora picks up from its last save on another machine, without doing anything twice.

Cora never writes mission documents directly. It asks the platform for every change, and the platform checks that the change is allowed. Cora’s rough notes stay in a scratch area that is not part of the mission.

Every run counts toward your organisation’s AI usage. When the plan’s AI usage is used up, Cora pauses her runs. See Managing usage.

Cora has helpers for work she hands off. None of them can change anything: they read the mission’s files and report back. Anything that is saved, posted or sent still goes through Cora’s own tools, and those wait for a person to approve.

  • The drafter writes long drafts, such as a plan, an email or a summary, in its own space. Its reading of the mission files and its rework stay out of Cora’s conversation, but the draft text itself comes back to her. Cora saves it herself.
  • The reviewer checks a draft before Cora shares or saves it. It checks the draft against the mission’s objectives and documents, the organisation’s and mission’s CORA.md rules, any style guide in the documents, and anything that needs a source, such as an offer, a price, a finance or legal statement, a statistic or a comparison. It also applies the Australian Consumer Law and Spam Act guardrails and the skills switched on for the mission. It answers with a verdict, pass or needs changes, and a list of issues. Each issue has a severity (blocker, major or minor), where it is, what is wrong and the fix.
  • The general-purpose helper does quick lookups in the mission files. It runs on a smaller, cheaper model.
  • The researcher finds facts and says where each came from.

Cora runs the reviewer on any draft that will go to people outside the mission or makes a claim. She fixes every blocker and major issue, or tells you plainly which ones are still open. She only calls a draft checked when the reviewer returned a clear pass in its structured result; any other answer means the draft is unchecked, and she runs the reviewer again or tells you so. She saves the draft once to a scratch file so the reviewer can read it there. A pass is not a person’s approval: you still approve what she saves or sends.

The reviewer cannot read Cora’s memory itself, so Cora gives it the CORA.md rules that apply. Each helper’s work shows separately in your organisation’s usage.

Cora keeps one short, structured status for each mission, and updates it herself. It shows where the plan stands. It has four parts:

  • Health: On track, At risk or Off track. She never marks a mission On track while something is overdue or blocked.
  • Stage: where the mission is in its life: intake, requirements, plan, tasks, delivery, review or closed.
  • Next milestone: a short title (up to 120 characters) and a date, when there is one.
  • Summary: one line of up to 200 characters.

She sets it with her daily status update, and again whenever the health, the stage or the next milestone changes. The platform keeps the latest status and every earlier one, and writes each change in the organisation’s history as project.status_set. Home shows the latest status of each of your active missions (“No status yet” until she has set one). For the Kestrel Cycles spring mission, it might read: At risk, Plan stage, next milestone “Objectives agreed” on Wednesday 12 November, “Waiting on the owner of the offer to check the objectives.”

The tool is called set_project_status. It can only change the status of the mission Cora’s run belongs to. It cannot name a person or another mission.

Cora can do work at a set time or on a repeat when a member asks her in the mission: “every weekday at 9 am, post yesterday’s open tasks”. She keeps each one as a schedule, with the tools create_schedule, list_schedules, set_schedule_state and cancel_schedule, and she is told the current date and time at the start of every piece of work. See Schedules for the limits and who can pause or delete one.

Mission members can open any document Cora writes for the mission, such as brief.md, in the Artifacts tab of mission details. The viewer shows text only: it does not run anything in a document, and you cannot edit a document there. Since M3, you can also see each version of a document and who wrote it. See Documents.

  • D0851: a mission’s goal is its objectives, and Cora is its mission coordinator.
  • D0122: any member, not only an org admin, creates missions.
  • D0126: each mission has one dialogue. D0545 amends it: people now post in it in their own name, as typed or polished with Cora.
  • D0118: skill packs are pinned to a version per mission, so a pack update never changes a running mission.
  • D0113: the platform is the only writer of mission records and documents; Cora asks it for every change.
  • D0852: mission admins change a mission’s name, objectives and colour.
  • D0623: a mission’s admins add and remove its people.
  • D0134: every model call is metered, and Cora’s runs pause when the AI usage is used up.