Agents · Claude Code

Run Claude Code in the background

Hand Claude Code a task and close the tab. wisp runs it headless in its own git worktree, routes its questions and permission prompts to you, and hands back a one-screen report with the real diff and the test run.

Claude Code docs

Free and open source · uses your own Anthropic account · macOS, Linux, Windows (beta) · Node 20+

A wisp report for a task Claude Code finished: headline, numbers, the diff and the test run

What wisp adds to Claude Code

A worktree per task. Each task gets its own branch from origin/main, so several Claude Code runs can work on the same repo at once without colliding.
A report instead of a transcript. When the run ends you read a headline, the numbers that changed, the real diff from its own edits, the test run and what to do next.
Questions mid-run. When it needs a decision, the card moves to Needs you and you get a notification, on your desktop or your phone.
Ship it. One click commits, pushes and opens the pull request; the card follows CI and review, and Merge syncs your local main.
Take over any time. Press T and your terminal opens in the worktree with claude --resume <id>.

How wisp runs Claude Code

wisp doesn’t replace or wrap Claude Code’s model access: it starts the claude CLI you already have installed and signed in, in headless mode, and reads its event stream. Roughly:

claude -p --output-format stream-json \
  --permission-mode acceptEdits \
  --append-system-prompt "<wisp protocol>" \
  --mcp-config '{"mcpServers":{"wisp":…}}' --verbose

Follow-ups resume the same session, so Claude Code keeps its context. Override the binary with WISP_CLAUDE_BIN.

Permission modes

Pick a mode per launch in the launcher (ShiftTab cycles them) or with --mode on the CLI. Here is what each one passes to claude:

wisp mode	Claude Code flag
Plan (read-only)	`--permission-mode plan`
Edits	`--permission-mode acceptEdits`
Full auto	`--permission-mode bypassPermissions`

In Edits mode, Claude Code’s permission prompts go to wisp’s approve tool (--permission-prompt-tool). The card asks Allow Claude Code to run npm publish? with Allow, Always allow in this task and Deny.

Questions, `/btw` and take over

Asking you: An MCP tool, mcp__wisp__ask_user, added with --mcp-config. The run pauses, you get a notification, and your answer goes straight back into the same session.
/btw side questions: Type /btw did you update the docs too? while it works. The question is answered on a copy of the conversation (--resume <id> --fork-session), so it knows what the run did and why; the running task never sees it.
Take over: Press T on the card, or run claude --resume <id> yourself in the task’s worktree.
Models: Opus, Sonnet or Haiku (or whatever Claude Code is set to), plus a thinking level passed as --effort.

Several folders in one task: the others are passed as --add-dir.
Claude in Chrome: when it is on by default in your Claude settings, wisp adds --chrome, which headless runs need.
Subagents: a Claude Code lead can delegate work to Codex, opencode or another Claude through wisp’s delegate tool.

Get started

Install Claude Code and sign in. Follow Anthropic’s instructions until claude runs in your terminal.
Start wisp.
Pick Claude Code in the launcher, describe the task and press Enter. From a shell: wisp "fix the flaky auth test" --agent claude.

Run it next to other agents

The same queue takes every harness. Compare one task on two of them side by side, or let Auto pick the harness and model per task from what worked before.

Guides: Run Claude Code and Codex in parallel · Run coding agents on multiple machines · Multi-agent coding across Claude Code, Codex and more.