8.3 — Watching progress

8.3 — Watching progress

Once your coding agent is connected and the first AC has been claimed, the Track tab on the Development page is where you live. Three surfaces give you real-time visibility into what's happening, each at a different zoom level.

Kanban board mid-run

1. The kanban board (mid-zoom)

The board groups acceptance criteria into columns by status:

  • Ready — claimable. Marked ready by you (from draft), hard prereqs all passed.
  • In Progress — claimed by an agent (via MCP) or a user, work underway.
  • In Review — submitted with a git SHA and diff URL, waiting for human (or CI) approval. Agents can't resolve their own reviews.
  • Passed — approved. Unblocks any AC that hard-depends on it.
  • Failed / Blocked — needs attention. Failed ACs go back to Ready when you re-open them; blocked ACs carry a category + reason and return to their prior status when unblocked.

Cards animate as their status changes, so even without polling for updates you'll see motion as your agent works. If you've planned sprints, the board groups cards into per-sprint sections above the unsprinted/Backlog group, with a Foundation section at the top for kind: "bootstrap" ACs (the scaffolding for your auto-derived Foundation stack).

2. The Activity feed (narrative zoom)

Below the kanban there's a collapsible Activity panel. Click to expand. It shows a chronological feed of every kanban transition with a human-readable formatting:

Agent vm_a1b2c3 claimed User can sign up with email · 2m ago Agent vm_a1b2c3 submitted for review User can sign up with email · 18m ago You approved User can sign up with email · 19m ago

The feed polls the events endpoint every 5 seconds (and on demand via the refresh button in the panel header). When new events arrive while the feed is collapsed, an N new badge appears beside the header — click to expand and read.

Use the feed when you want:

  • A timeline of one agent's session — scroll through to see the sequence of claims/submits/approvals.
  • To spot blocked work — blocked events show as rose-colored rows with the "blocked" verb.
  • To audit who did what — every row names the actor (You, Agent vm_xxxxx, CI, System).

The feed caps at the most recent 100 events to keep the DOM light. Older history is still in the database (kanban_events table) and queryable via the API.

3. The tab status dots (high zoom)

The Development page's tab strip — Plan, Hand off, Track, each with a small status dot — is your at-a-glance "is this still set up right?" reminder, and the readiness checklist covers what each dot asserts. If your API key gets revoked or your spec gets reset, the relevant indicator flips from green back to pending.


Tips for unattended runs

VibeMap's MCP loop is designed to let your agent run for an hour or more without supervision. A few patterns that make this safer:

  • Check in every 5–10 minutes, not after every AC. Approve in batches. The agent will keep moving as long as it has claimable work.
  • Watch the Activity feed for stalls. If you see no new events for 5+ minutes during an active run, your agent has probably crashed, lost connection, or hit a problem worth investigating in its own logs.
  • Use the Sprint Gantt as a rate gauge. The "passed/total Criteria" totals in the Gantt's bottom row update live — you can see the rate at which ACs are completing across sprints.
  • Don't ignore the Backlog. If ACs are landing in Backlog instead of the planned sprints (visible at the bottom of the Track tab), it usually means the LLM hallucinated some slugs during planning. Re-plan, or accept it and drag them manually.

Why there's no "Pause / Resume" button

The coding agent runs in your IDE, not in VibeMap. We can't stop or start it from the web app. To pause a run, just tell your agent to stop in its chat or close the IDE. To resume, give the same "get next ready and implement" instruction again — claims persist, so a previously-claimed AC will still be reserved for you.

If you want stronger control, you can revoke the API key from the Account → Developer tab. Any subsequent MCP call fails authentication and the agent stops cleanly.