Mission Control

Overview

Mission Control is the screen you land on after signing in — route /. It's a single, reshaping screen rather than a set of separate "new mission" and "results" pages: when nothing is happening it's a command center built around one big text box; the moment you submit a request it becomes a live view of the swarm working, then settles into a finished report.

Starting a mission

Type what you want done into the hero box, in plain English — for example "Audit my S3 buckets for public access." A few things about the box itself:

• Typing / opens a slash-command palette (the same one behind ⌘K) with shortcuts like /run and /teach; arrow keys navigate it, Tab / Enter selects, Esc closes it.
• Enter submits your request; Shift+Enter adds a new line. The box grows as you type, up to a limit, then scrolls.
• A colored dot to the left of the box shows its state: gray when empty, a green glow once you've typed something, a pulsing brand-colored dot with "Planning mission..." while it's being planned, and green again once ready.
• Don't want to type your own? Click one of the four example prompt cards to fill and submit instantly — things like "Provision a staging Redis cluster on AWS" or "Run a cloud cost audit for the last 30 days."

The Pre-Flight Manifest

Before anything is dispatched to the swarm, SnapFlow always shows a confirmation screen — the Pre-Flight Manifest — quoting your exact request back to you along with what's about to happen and why.

At the top, three chips summarize the request at a glance:

Below the chips, a wallet preview shows your credit balance before and after the mission (in red if it would go negative), then the single most important line — the verdict:

Underneath, you may see contextual notices: a 🛡️ note that a critical floor is engaged (certain actions always require approval, no matter your mode — see the critical floor), a ⚙️ note if the mode was automatically tightened for this specific mission (for example because the target is production), a ⚡ simulated-mode notice with a "Connect a provider" link, and 🔐 credential rows — if the request needs a new service's access, a "Generate & store" button provisions it into the encrypted vault right there, so you never type a password into a chat box.

A checklist below shows pass/fail items (✅/⚠️/❌) — credit balance, any blocking security-rule violations (marked BLOCKING), and a collapsible list of pre-existing, non-blocking posture findings about your cloud setup unrelated to this specific request.

Finally, two buttons: the primary action (label follows the verdict, e.g. "⚠️ Request approval" if a manager must sign off first, or "Blocked by security gate", disabled, if it can't proceed) and Cancel, which closes the manifest without submitting anything.

The missions list

Beneath the intent box, a running ledger of your missions, most relevant first. The lead card is always whichever mission is currently running, awaiting your decision, or — if neither — the most recent one, shown expanded with its full step-by-step ledger. Everything older collapses into an "Earlier" section as compact rows, color-tinted by outcome: green for verified success, red for failed or cancelled, orange for unverified or needing you.

Each row shows your original request, a one-line status sentence, and a status chip. Clicking a row opens the full Mission Run detail in a slide-over panel. Rows waiting on a decision keep Approve / Reject buttons right there in the list, so you never have to open a mission just to act on it.

Watching a mission run

This is the canonical "what's happening / what happened" panel for one mission — open by clicking any row in the missions list, or automatically shown while a mission you just started is executing.

Mission header

A pulsing colored orb (or, once settled, a glyph — ✓ verified, ✕ failed, ▲ needs you) shows live state next to "Mission · [elapsed time]" and your original request text. A status chip sits on the right; see the status vocabulary below for what each one means. Once the mission is finished, a Dismiss button clears it from the active view (it stays in the missions list).

Stage stepper

A horizontal tracker — Plan → Execute (or "Approve" while gated) → Verify → Result — with each node showing ○ upcoming, ● active, ✓ done, ▲ needs approval, ✕ failed, or ◆ verifying. The connecting line fills in green as steps complete.

Step ledger

One row per agent/step: a colored status glyph, the agent's name (each agent has its own color), the plain-language instruction it was given, and a status word — Queued, Planning, Running, Done, Failed, or Skipped. Click any row to open that step's own Task detail page.

Live activity (the flight log)

A collapsible feed of everything the swarm is doing, in real time — this is the "flight log." Two view modes, toggled with buttons:

Each line is tagged by lane — PLAN, LLM, TOOL, SYS — with its own color, alongside the acting agent's name/color and a timestamp.

When a mission needs your approval

If a step is gated, the Mission Run panel shows a Gate Evidence card in place of the live feed: the specific action, the target environment (tagged prod in red if it's production), why it was gated, which resources are touched, a Blast radius list of anything else that would be affected, the estimated cost, and a written rollback plan.

Neither button commits on the first click — SnapFlow always asks you to confirm before dispatching real work or discarding a plan. For everything else about approvals — including approving from your phone via a shared link, and the four permission modes that decide what needs your sign-off in the first place — see Approvals & permission modes.

When a result can't be auto-verified

Occasionally a mission finishes but SnapFlow's automated grader can't confirm the outcome. You'll see: "Work completed, but automated verification is unavailable. Review the steps above and confirm." with two buttons — Verify & charge and Reject & refund — each behind its own confirm step ("Verify and charge?" / "Reject and refund?").

Cancelling a mission

While a mission is still running (and not already paused for approval), a Cancel mission button is available — it reads "Cancel & roll back" if the mission has already failed or is blocked. This opens a modal tagged Destructive:

1. A warning explains that cancelling stops everything immediately, tears down any cloud resources the mission already created (so nothing is left half-built), and refunds any credits held in escrow.
2. An optional Reason field, logged for later reference — placeholder text: "e.g. wedged on a missing permission".
3. A required checkbox: "I understand this stops the mission and destroys any resources it created."
4. Click Cancel & roll back (disabled until the checkbox is ticked) — or Keep running to back out.

Outcome, resources, and the mission report

Once resources exist, a card lists how many were provisioned along with a verification badge — Verified by proof, Verified, or Human-verified — and each resource's cloud service, ID, region, and an Open ↗ link straight into your cloud provider's console.

The Mission Report is the readable write-up: a colored verdict line (green = good outcome, orange = issues found), collapsible sections mirroring the agent's own write-up (tables, lists, and code preserved, with the first two sections open by default), and — if the agent's findings suggest follow-up work — a Recommended actions card. Each recommendation shows an impact chip (Read-only Changes infra Deletes resources) and a button (Plan it →, Fix this →, or Run again) that — after a confirm step quoting the exact new request — starts a brand-new mission through the same Pre-Flight Manifest. SnapFlow never re-runs completed work silently.

Once billed, a settlement line shows the outcome (charged / refunded / partial charge) with a "View receipt ↗" link to a formatted HTML receipt, and — right after settlement — a brief full-screen splash confirms whether credits were used or refunded, with a link to download the receipt. It dismisses itself after 30 seconds, or you can close it.

"How to fix" boxes for missing access

If a step fails because SnapFlow doesn't have the access it needs, an inline fix-it box appears right where the failure happened:

The first time a task hits a credential wall, a blocking "Manual Action Required" modal can also appear automatically, with the raw error and buttons for Open Cloud Connections or Dismiss (dismissing it is remembered, so it won't nag you again for the same task).

Notifications

Small toast pop-ups confirm things as they happen: your request was dispatched, a mode reminder ("Read-Only mode — mutating steps run as read-only analysis. Toggle to Write-Enabled to execute changes."), a simulated-mode reminder, a task failing, a rejection, or an approval expiring.

Sharing a result

From a finished task's detail page, click 🔗 Share result to generate a public, no-login link and copy it to your clipboard automatically (the button briefly turns into "✅ Link copied!"). Anyone with that link sees a clean, read-only summary card — the agent's name and status, a verdict line, key metrics, findings and warnings, a timestamp, and buttons to send it to 💬 Slack, 💼 Teams, or 🔗 Copy Link — plus a closing invitation to try SnapFlow themselves. It's meant for pasting into a chat channel, not for giving someone access to your workspace.

Task detail

Where a "mission" is the whole multi-step operation, a task is one step of it, run by one agent. Open a task by clicking its row in the step ledger, or from an approvals/notification link.

Top to bottom:

• ← Tasks back link, then a header with the agent's icon/name, the word "Task", the task's ID, a status chip, and a meta grid (Created, Completed, Duration, Client).
• Approval gate card (only while awaiting_approval) — an optional rejection-notes field plus ✓ Approve / ✕ Reject buttons, each with its own confirm step.
• Progress bar while active — a segmented bar, one block per step, colored by state (green done, blue/pulsing running, orange awaiting approval, red failed/rejected, gray pending), with a label like "Step 2 of 5 · Executing — [description]". If SnapFlow doesn't yet know the full plan, it shows an honest indeterminate shimmering bar labeled "Working…" rather than a fake percentage.
• Result — the narrative outcome, key metrics, and findings, or a structured artifact card.
• Cloud resources — any resources this task created: service, ID, region, health status, and an Open ↗ console link.
• Prompt — the original instruction given to this task, expandable if long.
• 🔗 Share result and 📄 Export Audit Report — once the task is done (see Sharing a result above).
• Live Execution Logs — streamed in real time while the task is active.
• Diagnostics (collapsed by default) — Verifier History (each automated check: pass/fail, policy name, confidence %, timestamp, stated reason) and Execution Trace (every message in the run, expandable to full tool input/output).
• Error Log — raw error text, only if the task failed.
• Approval History — past decisions on this task, with status, timestamp, and reviewer notes.
• Raw input payload — a collapsed JSON dump of exactly what the task was launched with.

When a mission pauses on something only you can fix

Sometimes a mission hits a wall that is yours to clear: a cloud role missing a permission, a GitHub App whose access was revoked, an expired token, or a connected server that has gone offline. When that happens the mission pauses. It does not fail.

The Mission Run panel shows a Paused — waiting on you card explaining what is blocking it, and where SnapFlow can be specific (AWS, GCP and Azure permission denials) a remediation guide with the exact permissions needed, a deep link into the provider console, and a ready-to-paste least-privilege policy.

You do not have to tell SnapFlow when you have fixed it. It re-checks by itself roughly every 30 seconds and resumes the moment the blocker clears — for whichever provider the block came from, not just AWS. If you would rather not wait, the card has an "I've fixed it — re-check now" button that answers immediately and tells you what is still outstanding if it is not clear yet.

Two things end a pause other than fixing it: you can Give up & roll back at any time, and a mission left paused for 24 hours stops on its own and settles, so credits are never held indefinitely by something nobody intends to resume.

When an agent runs out of time

Agents work to a time budget. An agent that reaches it mid-task has not failed — it has run out of clock partway through real work — so SnapFlow hands the remaining work to a fresh agent rather than marking the step failed.

The hand-over is not a blind retry. After each cycle a separate assessor looks at what was actually achieved and estimates how much is left, producing a concrete list of outstanding items that becomes the next agent's brief — so it continues the work instead of starting over, and can be handed to a different specialist if the work has changed shape. If the assessor judges a different kind of expertise is now needed, the mission re-points the step accordingly.

Status vocabulary

SnapFlow uses one consistent set of states across missions, steps, tasks, and connections. Learn these once and they apply everywhere: