Looprch

Looprch runs a SEV3 project specification phase by phase with coding agents. SEV3 defines the project; Looprch coordinates the Planner, Plan Debater, Implementer, Tester and Reviewer that build and verify it, checks machine evidence for every gate, and keeps the execution state in plain files inside your project.

Status: 0.8.4, on npm as looprch. Linux and macOS (Windows through WSL).

Quickstart

npx looprch@latest install                 # or: npx github:NourHayik/Looprch install
cd my-project                              # SEV3 package (phases/, requirements/) at the root
looprch add . --agents cursor,codex,opencode --yes

Then, inside your agent:

/lr-init      # readiness, SEV3 trust, gate acknowledgement, roles and models
/lr-phase     # one phase: plan, debate, implement, test, review, handover, merge
/lr-auto      # all phases until the closure phase
/lr-finish    # closure phase and final report

From any terminal: looprch status, looprch log, looprch doctor.

How it works

The looprch CLI owns a deterministic state machine. The agent you talk to (the Lead) asks looprch next for exactly one action, performs it (start a subagent, dispatch a relay, run the gates, commit a checkpoint, ask you a question) and reports back. A new chat, another agent or a reboot simply continue. Roles run either as the host agent's own subagent (Direct) or through another agent's CLI with delegate-skills (Delegate). Every phase runs on a branch looprch/P-NNN and is merged and tagged when it closes.

Documentation

• User guide: docs/user/index.md
• Developer guide: docs/dev/architecture.md
• Contributing: AGENTS.md, docs/dev/contributing.md

License

MIT