claude-agent-map: see every agent, and what it’s doing
A short note on claude-agent-map: a Claude Desktop pane that shows the main session and every subagent — tier, time, tokens, and tasks — so parallel agent runs stay readable.
Superlite will happily put a dozen agents to work at once. That is the point. It is also the problem.
When the build is going well, the transcript is a waterfall of launches, reviews, and checkpoints. When it is not, you need to know which agent stalled, which task it owned, and whether the coordinator is still thinking or just waiting. Grepping progress.md and metrics.jsonl works. It does not feel like watching the run.
So I split out claude-agent-map: a small Claude Desktop plugin that opens a read-only Agent map pane beside the transcript.
What it does
It shows the main session at the top, and every subagent under the agent that spawned it. For each one you get:
- Who it is — type and short description
- What it is working on — superlite task ids when those records are present (✓ once checkpointed)
- How it is doing — green while running, red if aborted or blocked, grey when done
- What it cost so far — tier, elapsed time and tokens, plus a detail view for model, parent, ledger lines, and last message
The pane only reads. It never steers a run.
Open it with /claude-agent-map, optionally pointing at a project (or its .superlite folder). You can also ask Claude to open it, or reply approve map when superlite offers the pane at plan approval. Installing superlite from the marketplace pulls this plugin in as a dependency.
Why it’s useful
Parallel agent workflows fail quietly in the UI long before they fail in the tests. Eighteen agents is not “visibility.” It is a pile of IDs. The map turns that pile into a picture: who spawned whom, which tasks are still open, and which nodes went red.
Just enough structure that you can leave a swarm running and still answer “what is happening right now?” without opening six files.
The repo is here: https://github.com/iha2/claude-agent-map.