Tasks (markdown outliner, in development) (part 8)
Any task can host its own Claude agent chat. On a focused outliner row, the "Launch agent" button or the g key runs open-or-launch: if the task already has a chat, it reveals in the Sessions tab and selects it — free, no spawn, no pop-out — with a toast saying so (the task already has an agent session from its start date, named, so no new agent was started; a plain version without the date and name shows when that session cannot be loaded). If it doesn't, an editable launch-prompt popup opens, showing the visible task message the agent starts from (the task text + its open subtasks) plus a session-type picker — nothing is spent until you press Launch, and launching never makes a duplicate: the new session is attached to the task for you.
On this page
What it is
This is part 8 of the Tasks (markdown outliner, in development) page. It carries the next stretch of the material on that page, moved here because a single page is capped at 40,000 characters.
Where to find it
Reach this part through Tasks (markdown outliner, in development) — it lists every part and explains where the feature lives in the product. Everything below is reached from the same place.
How it behaves
Everything below is the behaviour, detail and edge cases that belong to this stretch of the Tasks (markdown outliner, in development) page.
Start a session — the welcome screen and its cards
On desktop, the Sessions tab's left rail is just your session list. When no session is open, the big panel to its right shows the Task Manager welcome screen: the swirling Omniscio orb, "Welcome to your Task Manager." in the brand gradient, the line "Choose an option below to get started.", and the cards. Start a session sits alone, centered, at the top; the six guided cards sit below it in a grid of at most three columns — three across and two down when the panel is wide, fewer columns when it is narrow — so the cards never squeeze. Open a session and its chat fills the panel; close it and the welcome screen comes back. It shows even when the rail is collapsed.
- Start a session (the card on top) opens a blank Tasks session right in the panel, so you can ask anything. It is the same session the rail's + New and Ctrl+T start, opened the same way, inside Tasks, and it costs nothing until you send your first message. It is the only card with a soft cyan outline, the same treatment as Mission Control's Start a session, because it is the one card with no pre-written prompt.
- The six guided cards — Plan My Day, Brain Dump, Clarify Tasks, then End My Week, Free Up Time, and Tame the Overwhelm. On a wide panel Plan My Day sits directly above End My Week, the day and week bookends. Click any of them and a session opens and greets you with its own first questions — no picker, no prompt to type. It's the fast way in when you don't have a specific task in hand: you just pick the kind of session you want and start talking.
- Every card runs on your own AI. A quiet line under the cards says All sessions default to using your selected AI model. When you have no AI account connected (no Claude login, no API key, no ChatGPT/Codex account) and no Omniscio credits, every card is grayed out and can't be clicked, hovering one says why, and the line adds Connect an AI account or add credits to start one with a Set up AI link to the AI account settings. While the app is still loading your accounts, the cards stay clickable rather than flashing locked, and a credit balance that can't be read leaves them clickable too, because starting a session on zero credit is refused with the Add credit prompt anyway.
- The card icons share one gradient. A single lavender-to-cyan layer — the same colours as the welcome headline — is laid across every icon at once, top-left to bottom-right, so each icon carries its own slice of it and no two look alike. It follows the light and dark themes like the headline does.
On a phone the Sessions screen is the list itself, so the six guided cards sit above the list there in one column, as before, with the gradient running top to bottom.
Plan My Day starts the same Plan my day session as the Plan my day with AI button on Today: a session that reads your plate and priorities and plans the day with you. It runs on your own AI like any session, and it asks to turn Plan my day on first if it is off. In English, the other five cards cost nothing until you reply. Each uses the same deferred pattern as AI Coaching: a pre-authored opening message shows the moment you click — no model is spawned and no tokens are spent yet — and the model only cold-spawns when you send your first reply. So opening one, reading its opener, and closing it is free. That opener is written in English, so it is shown only when the interface language is English (any region, such as en-GB). In any other interface language the card skips it: the session starts at once and the AI writes its own first message in your language, so that first message does use your AI (resolveMacroSessionDeferred returns no opener and macroSessionLaunchPayload launches the steering prompt immediately). Back at the top left of the session returns you to the cards (Escape does the same); the session keeps running. Only the Mission Control and Tasks card sessions have this Back button; other sessions do not, and neither does a popped-out session window, which has nowhere to go back to. Under the hood each launcher is seeded with a hidden super prompt that steers the whole conversation; you see the opener and your own messages, never the steering. Like every task agent, these run in the hidden Tasks workspace (__tasks_workspace__) and land in the Sessions list in the background.
The presets are built-in super prompts you can edit — they live in the bundled Super Prompt library as macro-brain-dump, macro-clarify-tasks, macro-overwhelm and macro-weekly-review (plus the shared eliminate-automate-delegate-opportunity-helper prompt behind Free Up Time), each carrying the hidden steering body plus its pre-authored opener; Plan My Day is not a preset: it carries no super prompt and starts the plate-bound Plan my day session instead. Still behind the Tasks Labs flag.
Plan — chat your way to a plan (removed)
The panel's tab strip once carried a third tab, Plan, beside Tasks and Sessions. It has been removed: the strip shows Tasks and Sessions only, and nothing in the app reaches a plan chat now.
Break it down — a session with an agent (M4)
When a task feels big or vague, Break it down helps you split it into next steps. The row's ⋯ menu action, the b key, the Break this down? hint, the Break it down button on a put-off task in Today and a breakdown tip all start a Break It Down with an Agent session for that task (the same type as picking Break it down in the Launch agent popup), then switch the Tasks panel to its Sessions tab with that chat open. The agent asks about the task first, proposes concrete steps, and offers to adjust them or create them as subtasks. It runs on your selected AI model and uses AI on every turn. If the task already has a running agent, nothing new starts and a message says so.
The old in-panel interview is still in the app, but no button opens it now. Its engine answers the command line (GET /tasks-v2/interview, POST /tasks-v2/interview/run, /send, /accept), and the rest of this section describes that engine:
- It's an interview, not a spawn. It uses the same cheap, bounded AI as the "Right now" / daily-check-in prioritizer (Haiku via the embedded key, short bounded prompts, defensive parsing). The model never sees full task bodies — only a bounded brief (the task text, its list's master context, a few attached-context labels, the existing sub-steps so it won't re-propose them, and the Preferred task size from About you, so a "coarse" planner gets a few big chunks instead of the default 3-7 small steps). Each reply also carries only the most recent 12 messages of the interview, so in a longer breakdown anything said before those is no longer sent — repeat anything from the start that still matters.
- Starting it spends one bounded call; reopening an existing breakdown is free (the thread is saved). No daily budget limits it: every turn runs on your Claude account first, then on your saved keys or credits. When the built-in AI helpers need an API key of your own, Break it down says so in the same words as the daily check-in: opening it shows that sentence with a link to the API key setup under Settings → Accounts, and a later turn answers with the same sentence instead of a generic "unavailable" reply. A key that was rejected is its own case: an expired or revoked key is never reported as a temporary outage — Break it down and the daily check-in both say the key was refused and point you at the same key setting, because waiting a few minutes cannot fix a dead key.
- A degraded reply says it is degraded. When the AI does not run — it is off, or the model is unavailable — the check-in tells you its summary came from your task list only — that marker is saved on the check-in itself, so it still shows for the scheduled daily run and after a restart, and the scheduled run logs a warning each time it falls back — and the Today card's "Suggest" says it is showing your own priorities rather than a prioritizer pick. A degraded ranking is therefore never indistinguishable from a real one, and a retry after a rate limit tells you how long to wait instead of "too many requests". A check-in reply cut off for room is finished for you: Tasks asks once more and joins the parts; if still cut off, it shows the partial reply with a Continue button. AI arrange on the Today plate says why it fell back to your priorities: the arrange result carries
degradedReason(offwhen the "AI arrange for the day's plate" setting is switched off,errorwhen the model was unavailable or answered garbled), so the line under the toolbar reads "AI is off, so this uses your own task list" with a Turn AI back on button, or says the AI was unavailable. It is a politerole="status"line, never an alert, because the plate was still arranged. A genuinely failed arrange call (nothing arranged) is reported once, by the store's toast; the view adds no second message. - It proposes; you decide. The AI asks at most one short question at a time, then proposes a handful of concrete next-actions (usually 3–7, sized to your preferred task size if you have set one). They land in an editable list — reword, remove, or add your own. Nothing is written until you click "Add N subtasks", which creates them all at once under the task (via the normal task-create path, inheriting the task's list). That add is one undo step: Ctrl/Cmd+Z straight afterwards removes every step the breakdown created and puts the task's own estimate back the way it was, so the shortcut never reaches past the breakdown to an unrelated earlier edit. The AI can never add tasks on its own, and the write step is also available as a CLI route (
POST /tasks-v2/interview/acceptinai-routes.ts, mutation-gated viatasksMutationGate— feature-flag + rate-limit, not approval-gated;ai-checkin-routes.ts:162–186). - Your reply shows right away. When you send a reply, it appears in the thread at once with a typing bubble under it while the AI answers (
sendInterviewMessageechoes the turn into the store before the round trip). If the send fails, the echoed turn is taken back, the thread is re-read so a turn the server did save is not hidden, and your text returns to the reply box. - It's aware of the time-estimate field — but only fills it on request. The AI knows the estimated-time field exists and sees the task's current estimate (if any), so it can reason about how long the work takes. Only when you ask about time — "roughly how long will each step take?", "give me estimates", "how long is this in total?" — does it propose a per-step estimate and offer a running total; unprompted it stays silent on time (a task app shouldn't guess durations you didn't ask for). When it does estimate, each step shows a small ⏱ chip and the panel adds an honest total line ("≈ 1h 15m total · 2 of 3 steps estimated" — truthful about partial coverage). The line counts only steps with text, the same steps Add saves, and its total is capped at the 100-hour field maximum exactly as the saved parent estimate is (
rollUpEstimateinsrc/shared/tasks-v2-duration.ts, used by both the panel and the server). A breakdown holds at most 20 steps (INTERVIEW_MAX_SUBTASKS, the same cap the accept schema enforces): at 20, Add a step is disabled and the panel says why. On Add, each step's estimate is written onto its subtask, and if the parent task had no estimate yet and every step carries one, the steps' total rolls up onto it (with only some steps estimated the parent stays blank, because a partial total would understate the whole task) — so a big task finally shows a realistic total — but it never overwrites an estimate you set by hand. This is the "help me feel less overwhelmed" path: ask for a breakdown, then ask for the time, and you get both the steps and a grounded total. - Proactive nudge (free, default ON). "Looks-big nudges" (a toggle in the header's ⋯ More menu, the same label as in Customize) shows a subtle hint on rows that look big or vague — decided by a free local heuristic: a not-done task with no sub-steps that is (long text) or (importance High) or (has a deadline). The importance bar is intentionally High-only (stricter than the shared
HIGH_THRESHOLD) so the hint stays rare. No AI is called and nothing is spent until you actually click the hint.
Still behind the Tasks Labs flag. State for a task's breakdown lives in tasks_v2_interviews (the conversation + the current proposal + an "accepted" stamp); reopening never re-spends.
- A finished breakdown stays where you can see it. Once every step a breakdown added is done, the pane still shows the conversation, marked Done: every step is finished, instead of disappearing.
- Steps you already have are skipped, and you are told. When you add a breakdown, any step that matches a sub-step the task already has open is left out, and a note says so ("Skipped 1 step you already have.").
- An accepted breakdown can always be redone. Once you have added a breakdown's steps, the pane keeps showing that thread with its Added (or Done) mark and read-only steps — and under it a Break it down button. Pressing it first asks Start a new breakdown?; confirming retires the accepted thread and opens a fresh interview on the same task (the steps you already added stay where they are), so a first breakdown that turned out wrong, or one whose steps you later deleted, is never a dead end. Reopening a breakdown you have not accepted stays free and reuses the thread you already have (only the explicit redo, like starting one at all, spends a call).
- A damaged saved conversation is never written over. If a breakdown's saved conversation cannot be fully read (the stored text is not valid JSON, is not a list, or holds an entry that is not a message), sending the next message or starting the breakdown again stops with "Could not continue this breakdown because its saved conversation could not be read. Nothing was changed." The stored text is left exactly as it was, and both sending (including resending the same message after a failed try) and starting again refuse before the AI is called, so nothing is spent. This matches the daily check-in, which refuses the same way.
Tame the overwhelm
A coaching session type in the launch picker, Tame the overwhelm, runs The Overwhelm Coach — a warm, steady guide for when the pile feels bigger than the day. It works in two gears. First it gets you calm and clear: normalizes the feeling, offers a 60-second reset (never opening with the to-do list), then empties your head, cuts the list ruthlessly (delete / defer / delegate / renegotiate), and ranks the vital few (ICE, head-to-head). Then it gets tactical: it shapes an "Overwhelm Battle Plan" with a concrete Today's 1–3 and a trickle plan for the rest — and when a priority is itself a big project, it runs the Eat-the-Elephant engine, breaking it into a phased, bite-sized roadmap with an easy first win to start on today.
Unlike Clarify & enrich (project-launch only), it's offered on both a single task and a whole project — pick Tame the overwhelm from the Launch agent popup (the button / g on a task, or a project's "Launch agent…" menu). Like Interview for context and Clarify & enrich, it's a full background Claude session in the hidden Tasks workspace (real tokens per turn, cost-gated by the Launch button) — not the old cheap bounded Break it down interview engine. After you approve the plan, it files its concrete next-actions as real tasks via the Omniscio CLI — subtasks under the task on a single-task launch, or tasks in the project on a whole-project launch. When the plan includes an Eat the Elephant roadmap, its phases become subtasks with their steps nested beneath and the easy first win is flagged first. It never creates or changes anything before you approve.
Eliminate, automate, delegate
A coaching session type in the launch picker, Eliminate, automate, delegate, runs the EAD Analyzer — an operations-efficiency coach for getting work off your plate. It works the Eliminate → Automate → Delegate hierarchy in that strict order (the best task is one that doesn't need to happen at all): it interviews you to surface hidden work, then runs each item through the three lenses — Eliminate (zombie processes, the Two-Week Test, the "New CEO" test), Automate (only once a process is "perfectly defined"), and Delegate (who-not-how, delegation vs. dumping) — gently challenging every "but it has to be me". The deliverable is a categorized, prioritized, confidence-scored analysis of what to cut, automate, or hand off.
Like Tame the overwhelm, it's offered on both a single task and a whole project — pick Eliminate, automate, delegate from the Launch agent popup (the button / g on a task, or a project's "Launch agent…" menu). It's a full background Claude session in the hidden Tasks workspace (real tokens per turn, cost-gated by the Launch button) — not the old cheap bounded Break it down interview engine. After you approve the analysis, it files each recommendation's concrete next step as real tasks via the Omniscio CLI — subtasks under the task on a single-task launch, or tasks in the project on a whole-project launch — and, for a task it recommends eliminating, it checks that task off only with your sign-off. It can also save the whole analysis as a context note. It never creates, changes, or completes anything before you approve.
Interview for context
A sixth agent session type in the launch picker, Interview for context, adapts the "Context Handoff Generator": it interviews you about a task (or a whole project), writes one clean, paste-ready context document, and — only after you approve it — saves it as attached context that every future agent inherits. Two entry points, differing only in where the result is saved:
- On a task — the launch-agent picker's sixth option (the Launch agent button /
g, then pick Interview for context). On your approval it saves the document as a context note on that task (POST /tasks-v2/context/create), so every future agent launched on that task receives it automatically (via the same cascading-context brief the per-task chat gets). - On a project — the "Interview" button on a project's master-context card (beside its Edit button). It interviews about the whole project and, on your approval, writes the project's master context (
POST /tasks-v2/lists/update), inherited by every future agent launched on any task in that project. A confirm dialog gates the spend first.
Unlike the old Break it down interview engine (a cheap, bounded AI that only proposes subtasks and can never write on its own), Interview for context is a full background Claude session in the hidden Tasks workspace — it uses real tokens per turn and persists its output through the Omniscio CLI. It never hard-codes an id: the launch-time hidden footer hands the agent the concrete route + id + auto-trusted token for whatever it was launched on, and it saves the draft once you approve it (it is told to wait for your OK; Omniscio does not block an earlier save).
Clarify & enrich
A seventh agent session type, Clarify & enrich, is the whole-list groomer — it turns a messy, opaque task list into one an AI (or a stranger) could actually act on. Unlike the other types it is project-launch only: it never appears on a single-task launch (the picker filters it via LIST_ONLY_SESSION_TYPES), because its whole job is to walk the list, not one task. A single-task or batch launch that names it anyway — over the CLI or raw IPC — is refused, as is one naming Do it for me, which only ever runs through its own delegation flow.
Launch it from a project's "Launch agent…" menu (the generalized launch popup in list mode) and pick Clarify & enrich. The spawned agent walks the project's open tasks one at a time and, for each, shows its clean read and offers a three-way choice: A. Skip (already clear — leave it), B. Clarify, or C. Spawn. On Clarify it asks a few simple questions about a vague task, shows a before → after rewrite that preserves your wording and intent, and — only after you approve that task — rewrites the task's own text in place (POST /tasks-v2/update); where it helps it also offers to add a subtask (POST /tasks-v2/create) or attach a context note (POST /tasks-v2/context/create), but the description is the point. On Spawn it writes a ready-to-run prompt for the task (objective, context, definition of done, a do-it instruction) as a copyable block for you to launch a fresh session with — it never runs the task or launches a session itself (the same Spawn option Define & enrich offers). It handles an empty / already-clear list ("nothing to groom") and a project with more than 40 open tasks (the agent is given only the first 40, in list order, sub-tasks included; it grooms those and offers to continue) gracefully.
Like Interview for context, it's a full background Claude session in the hidden Tasks workspace (real tokens per turn, cost-gated by the Launch button) — not the old cheap bounded Break it down interview engine.
Per-task agent chat
Any task can have its own Claude agent chat. Highlight a task row and press its Launch agent button, or the g key. What happens next depends on whether the task already has a chat:
- It already has one. The chat opens in the Sessions tab and is selected. This is free: nothing new starts and nothing pops out. A message says so — the task already has an agent session from its start date, named, so no new agent was started (a plain version without the date and name shows when that session cannot be loaded) — and it sits in the default-on Tasks tips & hints toast category (Settings → Notifications → Tasks).
- It doesn't have one yet. A launch window opens. It shows the message the agent will start from: the task text and its open subtasks. You can edit that message. You also pick a session type: Custom, Break it down, Work through resistance, Tame the overwhelm, Eliminate, automate, delegate, Interview for context or Do it for me (draft).
- Where it runs. If you have real code projects, the window also lets you choose where the task runs. It starts at the code project chosen under "Run agents in" in this project's options menu, or, when none is chosen, in a hidden Tasks workspace. Pick a different project to run it inside that code instead. Your pick is for this launch only; the next launch starts from the project's "Run agents in" choice (or the workspace) again.
- See what else the agent gets. The agent also gets a hidden brief: coaching for the session type you picked, plus your project notes and any context attached to the task or the tasks above it. A folded Show what the agent will also receive line lets you read that extra text before you launch. It updates when you change the type, and it never shows an access token.
- Nothing is spent until you press Launch. The launch window is the cost check. No message is sent and no tokens are used until you click Launch. The agent then starts in the background and shows up in the Sessions tab.
- There is a launch limit. You can launch up to 30 task agents a minute and 200 a day. Every task-agent launch counts toward the same limit, whether you start it in the app or another agent or tool starts it for you. The daily count starts over at midnight your time, and a restart doesn't reset it. Past the daily limit, a launch is turned down with "Daily launch limit reached … please try again tomorrow." A launch another tool starts waits up to 16 minutes for the session to start, and one you start in the app itself waits a minute and then carries on in the background — so give a slow start time before retrying. A launch counts toward the limit as soon as it is accepted, and is given back if that start fails.
Whether you press Launch or Cancel, the keyboard goes straight back to your task list, so the arrow keys work right away. Coming back to the app window while Tasks is open does the same.
The agent starts already briefed. It gets the list's notes, context from the tasks above this one, the task text, its open subtasks, its attachments, a list of all your projects (with each one's open-task count, and its own project marked as current), and a short guide to the Tasks commands. With these it can read and update your tasks across your whole board, not just its own list: create, edit, complete, move a task to another project, or attach a note.
Unless the project has a "Run agents in" choice, the agent works in that hidden Tasks workspace folder. It is kept apart from your real code projects, so a task agent can't touch code you didn't point it at, and it never appears in the sidebar. If you pick a project in the launch window, Omniscio first checks that it is a real project you can start sessions in.
Launching never makes a duplicate. If the task already has an agent, that one is reused. If you edited the starting message, your text is sent exactly as written; if you didn't, the agent gets exactly what the window showed. The new session is also attached to the task for you, just as if you had used Attach a session, so it shows in the task's 📎 count (see Context attachments). Launching again never attaches it twice, and attaching can never block or break the launch. When you launch one agent for several selected tasks at once, it is attached to every one of them. One launch covers at most 25 selected tasks; with more selected, Tasks shows a hint instead of launching.
One agent for several selected tasks. If any selected task was deleted or archived before the launch went through, no agent starts: Omniscio names the tasks that are gone and asks you to select again, and nothing is spent. The CLI answers the same way, so a batch never quietly launches with only some of the tasks you picked.
gonly works when a row is highlighted and you are not typing in it. While you type,gis just the letter g. A full set of agent hotkeys, a lasting activity log and more visual polish are planned but not built yet.
Deleting a task also closes the chat it has, the same way closing it yourself would; filing it away leaves the chat running. Undoing the delete brings the task back but does not restart the chat, so start a new one if you want to carry on.
The Sessions tab — every task agent in one place
The left side of the Tasks panel has two tabs: Tasks and Sessions. Tasks is your task list, unchanged. Sessions turns that side into a list of your task agents. It works just like the Sessions tabs in KMS and Mind Map.
- A live count on the tab. When task agents are running (green), waiting on you (amber), or stuck with an error (red), a small colored number appears on the Sessions tab. On a phone it sits on the header's Sessions icon. So you can see that agents need you without leaving your list. Only counts above zero show, and the count hides while Focus Mode is on. Every Sessions tab in Omniscio (KMS, Mind Map, Flowchart, Whiteboard, SMS) shows the same badge, and it uses the same count as the sidebar and the Inbox, so the numbers always agree.
- Where task agents live. Every agent you launch from a task appears here, running in the background. The one exception is a task you launched into one of your real projects: it shows under that project's sessions instead. The task row's own open-agent button still reopens it.
- A sorted list. Agents are grouped into Needs You, live, paused, snoozed and archived. Right-click any of them to pause, archive, rename, pin or snooze it, just like in the main sidebar.
- Opening a chat. Click an agent to open its chat in this pane on a computer, or full screen on a phone. You stay in Tasks; nothing pops out. If the chat can't be loaded (for example, the session was deleted), a "Couldn't open that session's chat." message tells you so. On a phone you see a loading screen while it loads, and only a real failure shows "Couldn't load this task's chat." Its Retry button tries loading the chat again.
- Session hotkeys work on the open chat. While a chat is open here, E or Ctrl+W archives it (it stays on screen, grayed out, with an Undo message), P pauses it and H snoozes it. When you close the chat or go back to the Tasks tab, those keys work on your tasks again, so
eorxfinishes the highlighted task. - Start a fresh agent. The + New session button starts a new task agent that isn't tied to one task. It starts in the background and opens in this pane, just like the welcome screen's Start a session card. Ctrl+T anywhere on the Tasks screen does the same, and switches to the Sessions tab first.
On a phone there are no tabs, to save space. Tap the Sessions icon in the header instead (it carries the same live count). A ← Tasks button at the top of the list takes you back to your tasks. Like the rest of Tasks, this sits behind the Tasks Labs setting.
Related
The overview, the other parts, and everything else worth reading next all sit on Tasks (markdown outliner, in development).
Last verified 2026-10-26