Agents

Every coding agent CLI, one queue

wisp runs the agent CLIs you already pay for: Claude Code, Codex, Cursor CLI, opencode, Antigravity CLI and Grok Build. Each runs headless on its own branch, and each hands back the same one-screen report.

How they differ inside wisp

wisp gives every harness the same queue, reports, worktrees and shipping. What differs is what each CLI exposes: the flags behind each permission mode, whether it can load wisp’s MCP tools, and whether it can fork a conversation.

Claude CodeCodexCursor CLIopencodeAntigravity CLIGrok Build
CLIclaudecodexcursor-agentopencodeagygrok
Plan--permission-mode plan-c sandbox_mode="read-only"--mode plan--agent plan--mode plan--permission-mode plan
Full auto--permission-mode bypassPermissions--dangerously-bypass-approvals-and-sandbox--force--auto--dangerously-skip-permissions--permission-mode bypassPermissions
Asks you viaMCP toolMCP toolquestion blockMCP toolquestion blockMCP tool
/btw on a forkYesYesYesYestask recordtask record
Take overclaude --resume <id>codex resume <id>cursor-agent --resume <id>opencode --session <id>agy --conversation <id>grok --resume <id>

Adding a harness is one adapter file in server/adapters/.

The wisp launcher’s harness and model picker