Multi-agent engineering workflow. Replace hallucination with discipline.

Stop single-agent chaos. TMB runs your project like a real engineering team: a planner who scopes, an executor who builds, and an independent reviewer who catches what you'd miss.

The shipped product. Drop-in multi-agent workflow for any Claude Code project.

• bro — single Human entry point + planner. Discusses, designs, writes task specs, verifies SWE's work, closes tasks atomically. Never edits source.
• swe — executor. One task per spawn, isolated git worktree, atomic close. Cannot self-approve.
• pr-reviewer — independent 3rd party. Reviews diffs at push time. Read-only on files.

Plus: SQLite trajectory memory (resume across sessions, no codebase-rediscovery tax), hard-enforcement hooks (no prompt-discipline ceiling), 6-layer enforcement model from server-side middleware down to prompts. Detailed agent contract and enforcement matrix.

/plugin marketplace add trustmybot/plugin
/plugin install tmb@trustmybot

MIT-licensed, free forever. This will not change. Community PRs welcome.

A commercial tier for teams whose needs an MIT plugin alone can't promise.

Planned features:

• 🧠 Hosted trajectory memory — cross-session, cross-agent, cross-team
• 📊 Audit store — queryable history of every decision, with retention + export
• 🔒 Hard permission isolation — beyond what hooks can enforce
• 🌍 Cross-provider — OpenAI, Gemini, Groq, Mistral, DeepSeek, local models
• 🏢 Team features — SSO, audit export, dashboard, priority support

Boundary commitment: Enterprise will never gate features that make the plugin work standalone. The plugin is the open core; Enterprise adds infrastructure for teams.

If your team needs these and you'd like to influence what gets built first, open an issue or reach out to @ZaxShen.

Built by @ZaxShen.