@phamvuhoang/otto-core

Library half of Otto — a harness that drives the Claude Code CLI against a target repository in an iterating implementer → reviewer loop, running directly on the host OS.

This package is the engine: the iteration loop driver, the runner + NDJSON stream renderer, the prompt-template renderer, and the stage registry. The user-facing CLI lives in @phamvuhoang/otto (otto-afk / otto-ghafk).

Security: Otto runs Claude with --permission-mode bypassPermissions. Point it only at repositories and prompts you trust. See the repo's SECURITY.md.

Install

npm i @phamvuhoang/otto-core

Use

import {
  runAfk,
  runGhAfk,
  runLoop,
  STAGES,
  renderTemplate,
} from "@phamvuhoang/otto-core";

// Drive the plan/PRD loop from argv (same entry the otto-afk bin uses):
await runAfk(["<plan-and-prd>", "5"]);

Public surface: runAfk, runGhAfk, runLoop, STAGES, Stage, renderTemplate, runStage. Subpath exports: ./loop, ./runner, ./stages.

runStage spawns claude directly on the host with cwd set to the workspace directory. By default (OTTO_RUNNER=sandbox) it writes a transient --settings JSON that enables the native OS sandbox, confining writes to the workspace. Set OTTO_RUNNER=host to run unsandboxed. Credentials (~/.claude, ~/.config/gh) are read natively — no bind-mounts required. The templates/ directory (prompt playbooks) ships in the tarball.

Docs

Full usage, setup, environment variables, and architecture are in the main README and docs/ARCHITECTURE.md.

License

MIT © Henry Pham.