When you kick off a big workflow in Claude Code — the kind that fans out into dozens of sub-agents reviewing a diff or migrating a codebase — it feels like magic right up until it finishes. Then you’re left with a one-line summary and no real sense of what just happened. Which agents ran? How many tokens did the verify phase burn? What did agent #37 actually conclude? The answers are all sitting in ~/.claude, in a pile of JSON and JSONL files you were never meant to read by hand.

v0.42.0 reads that pile for you.

The problem

Claude Code’s Workflow tool writes a trail of artifacts as it runs: a summary per run, a directory of per-agent transcripts, a journal of events. It’s all there on disk. But “on disk” isn’t “visible.” A workflow that spawned 50 agents across two phases leaves you scrolling through opaque files trying to reconstruct what cost 3 million tokens. You can’t filter, you can’t compare runs, and you definitely can’t hand it to a teammate.

The solution

The Workflows tab is now a read-only dashboard for your dynamic workflow runs — no config, nothing to wire up. It scans ~/.claude and shows you:

A run list you can search and filter by status, project, model, and date.
Run detail — the phases (with real completion state), every agent with its token and tool counts, the saved workflow script, the final result, and a link straight back to the parent session.
Agent detail — each agent’s prompt, its result, and a bounded transcript of what it did.

It’s the difference between “the workflow finished” and “here’s exactly what the workflow did, agent by agent.”

There’s also a new Claude Home page: a safe, read-only browser for the rest of your ~/.claude directory — hooks, rules, jobs, and workflow definitions, with metadata and previews.

Trust over accuracy

The interesting design constraint here was secrets. Agent transcripts and workflow scripts routinely touch API keys, tokens, and credentials passed as tool arguments. Showing that raw in a UI — one you might screen-share or expose over a tunnel — is exactly the kind of thing that’s fine 99 times and a disaster the 100th.

So every preview these views render — scripts, results, transcripts, tool arguments, journal entries — passes through a redactor before it reaches your screen. It scrubs provider key prefixes, JWTs, bearer tokens, PEM blocks, and key: value credentials, while leaving normal prose readable. And the genuinely sensitive corners of ~/.claude — session-env, shell-snapshots, file-history — are shown as metadata only. We count the files; we never preview their contents.

The principle is the same one that runs through the whole product: better to show nothing than show something wrong. A leaked key or a wrong number costs more than a blank space ever does.

What’s next

Live run progress, so you can watch a workflow fan out in real time rather than only after it lands.
Cross-run comparison — token and phase trends across your workflow history.

Update now

npx claude-view@latest

Open the Workflows tab and pick your most recent run. If you’ve run anything ambitious lately, you’ll finally get to see what all those agents were up to.