Skip to content

v4.0.0 — reach: self-orchestrating agents

Choose a tag to compare

View all tags
@hanmariyang hanmariyang released this 18 Aug 14:48
· 73 commits to main since this release
v4.0.0
89ef8a5

Added

  • Self-orchestrating agents. Every local run can spawn its own sub-agents by writing .coxpit/spawn.json in its worktree — the daemon launches each subtask as an isolated sub-run and maintains .coxpit/subtasks.json with live status. A file, not an API, because default agent permissions (claude acceptEdits, codex workspace-write) block network calls — file writes are the one channel both providers allow. Orchestration moves inside the agent's own reasoning loop. Sub-runs carry ↳ by rN lineage on their cards. (COXPIT_AGENT_ORCH=0 disables.)
  • Start from GitHub. Paste an issue/PR URL — the task form drafts itself from its title and body (gh CLI for private repos, public API otherwise). You review, pick a provider, Run fleet.
  • Share a run. One click mints a read-only snapshot link — timeline + diff, server-rendered, no auth, no actions, revocable.

Upgrade: npm i -g coxpit (or desktop auto-update once assets land).

Assets 16
Loading