Bullswarm

Route bounded coding work across whichever installed agent CLI has quota to spare.

View the Project on GitHub cowcow02/bullswarm

Bullswarm is a CLI that sends a coding task to whichever of your installed agent CLIs — Claude Code, Codex, Grok, OpenCode, or Command Code — currently has unused subscription quota, then checks the result by its content.

It routes by quota and pace instead of a fixed map from work type to agent, and it judges every delegate’s output by content, never by exit code. A pool is one installed agent CLI, or one account of that CLI. A lane is the nature of the work: analyze (read-only), build (edits), or chore (mechanical edits).

Guides

Page What it covers
Entry points run vs workflow goal, and every top-level verb with its one-line purpose
Doctrine The four non-negotiable rules, and the result verdict shape
Getting started Install, integration with Codex/Claude/Grok, quick-start commands
Strategy Model/provider configuration, rungs (model plus reasoning level per effort tier), benchmark evidence
Workflows Authoring a program, kinds, advisories, the plan contract/validate/goal flow
Operations Listing and inspecting runs, the result envelope, context diet
Dashboard workflow watch, the interactive TUI, terminal glyph fallback
Repository operations The issue-watcher launchd agent
Routing How a pool is picked: pace, 5-hour headroom, expiring-soon urgency, load, quarantine
Providers The provider plugin contract: tiers, provider.mjs exports, the kit, the snapshot, every pool field and its reader

Internal notes and studies

These are historical working notes, audits, and experiment writeups, kept as records rather than as current documentation. Skip this list unless you are working on the Bullswarm codebase itself: