Watching progress
On this page
Watching progress
Real-time visibility into what your agent is doing — the Development board and the Activity feed. · Recommended
Once your agent is connected and the first AC is claimed, the Development page is where you live — it's Hand off to IDE in the project sidebar, or Sync with IDE on a code-first project. The board fills the whole page — there are no tabs — and two surfaces give you visibility at two different zoom levels.
Figure: The kanban board mid-run.
1. The Development board (mid-zoom)
The board groups acceptance criteria into columns by status:
| Column | Meaning |
|---|---|
| Blocked | Needs attention — carries a category and a reason. Sits far left so it's visible without scrolling. |
| Awaiting Approval | Still draft. Approve it to make it claimable. |
| Ready | Approved by you, prerequisites all passed — claimable. |
| In Progress | Claimed by an agent (via MCP) or a user; work underway. |
| In Review | Submitted with a git SHA + diff URL, awaiting your (or CI's) approval. Agents can't resolve their own work. |
| Done | Resolved. Passed ACs unblock anything that hard-depends on them; failed ACs land here too and can be dragged back to Ready. |
Cards animate as their status changes, so you see motion as the agent works. If you've planned sprints, the board groups cards into per-sprint sections, with a Foundation section at the top for the scaffolding ACs from your auto-derived Foundation stack.
2. The Activity feed (narrative zoom)
Below the board is a collapsible Activity feed — a chronological, human-readable log of every transition:
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 every few seconds (and on demand via its refresh button). When new events arrive while it's collapsed, an N new badge appears. Every row names the actor (You, Agent, CI, System), so it doubles as an audit trail.
Use the feed to:
- Follow one agent's session end to end.
- Spot blocked work — blocked events stand out in the feed.
- Watch for stalls — no new events for 5+ minutes during an active run usually means your agent has crashed or lost connection.
3. The Handoff panel (setup zoom)
Planning now lives on its own Sprint Planning page, so the Development page has no tab strip. Your connection details live one click away instead: the ⓘ Handoff Instructions button (top-right of the board, next to the Category filter) opens the Hand off to your IDE panel. It carries your API-key count with a link to Account → Developer, a one-line curl for the blueprint endpoint, and the MCP config block for your IDE. Reach for it whenever you need to re-connect an agent or check that your key is still live.
Reviewing and approving
When an agent submits an AC it lands in In Review. Approving is your job (or your CI's) — agents can't approve their own work. For each item:
- Read the diff the agent submitted (a git SHA and diff URL are required on every submission).
- Approve → the card moves to Done as passed, and the next AC in the chain becomes claimable.
- Send back → the card is marked failed (it stays in Done); drag it back to Ready and the agent picks it up on its next loop.
💡 Tip: Review in batches every 5–10 minutes, not after every single AC. The agent keeps moving as long as it has claimable work, and you avoid constant context-switching.
Tips for unattended runs
VibeMap's MCP loop is built to run for an hour or more without supervision:
- Check in every 5–10 minutes, approve in batches.
- Use the sprint Gantt totals as a rate gauge — the passed/total count updates live.
- Watch the Backlog. If ACs land in Backlog instead of planned sprints, the planner likely mismatched some slugs — re-plan, or drag them manually.
⚡ Power user: There's no Pause/Resume button because the agent runs in your IDE, not in VibeMap. To pause, tell your agent to stop or close the IDE; to resume, re-issue the "get next ready and implement" instruction — claims persist. For a hard stop, revoke the API key: the next MCP call fails auth and the agent halts cleanly.
↔ The traditional way
Traditionally, "how's it going?" is a standup question, a Slack ping, or a Jira board someone has to remember to update. Here the board is the work: your agent moves its own cards and leaves its own notes as it goes, so progress is something you glance at, not something you chase.
What's next
- Something not moving? → Troubleshooting
- Day-to-day board workflow → Kanban tracker (developer docs)
- Detect spec drift as the code evolves → Drift detection