BINGO
A shell-based framework that wraps Claude Code CLI to run autonomous, multi-step software development pipelines. Hand off a spec. Get back working code.
How Bingo Works
A two-phase pipeline that takes your specification and autonomously transforms it into working software.
Planner Agent
You provide a project spec — a text file or inline description. Bingo spins up a Claude agent with a planner prompt that reads your spec and produces a dependency-ordered task list (todo.md) where each item is small enough for a single agent pass.
Worker Loop
A loop picks up the first pending task, builds context (current phase, pending tasks, handoff notes, project skills), and sends it to a fresh Claude agent. The worker completes the task, marks it done, and the loop repeats until every task is checked off.
Built for Reliability
Bingo handles the hard parts of autonomous development — context limits, failures, and complex multi-step work.
Context Handoff
When the context window fills up, the session resets and context is handed off to a new agent via a handoff file — no work is lost.
Retry & Blocking
Failed tasks get up to 3 automatic retries. After that, the task is marked as blocked and the loop moves on to the next one.
Error Pit Stop
When a worker flags an unsolvable error, a separate error-fixer agent gets a fresh context to diagnose and fix the issue.
Session Continuity
Uses Claude's --resume flag to maintain conversation context across worker iterations, resetting only on phase transitions.
API Error Handling
Transient errors get exponential backoff retries. Rate limits trigger a cooldown period. Fatal errors stop the pipeline cleanly.
Skills System
SKILL.md files define project standards — design tokens, coding conventions, etc. Workers receive these as context and follow them.
Optimize Your Credits
Running Claude Code manually in a single long session wastes tokens on context rebuilding, retries, and bloated windows. Bingo's pipeline architecture keeps each agent focused, lean, and productive.
Fresh Context Per Task
Each worker starts with a clean, minimal context window — no accumulated bloat from previous tasks eating into your token budget.
Smart Context Handoff
When context limits approach, Bingo saves critical state to a handoff file and spawns a fresh agent. No tokens wasted hitting the ceiling.
Targeted Error Recovery
Instead of the main agent burning tokens retrying, a separate error-fixer agent gets a focused context to diagnose and resolve issues efficiently.
Stuck Detection
Bingo detects when an agent is looping without progress and intervenes — preventing runaway token consumption on unsolvable problems.
Tmux Monitor
Run via bingo_tmux.sh for a three-pane dashboard that shows everything happening in real-time.
Agent loop output — see every task as it's processed
Live Claude streaming — watch the AI think in real-time
Token usage and task progress — track resource consumption
Clean & Modular
Everything is organized into clear, single-purpose shell scripts. Templates define agent behavior, libraries handle utilities, and runtime state is isolated per-project.
Templates
System prompts that define how each agent type (planner, worker, error-fixer) behaves
Libraries
Shared utilities for parsing responses, managing todos, tracking tokens, and detecting stuck agents
Runtime State
.bingo/ directory keeps per-project logs, session data, and handoff files isolated
Get Started
Go from zero to autonomous development in under a minute.
bingo-init my-appbingo project_spec.txtbingo -d "Build a REST API with auth and CRUD endpoints"bingo-tmux project_spec.txt# Resume workers if you already have a todo.md
bingo --workers-only
# Check current status
bingo --status
# Start fresh
bingo --resetStop coding manually.
Let Bingo build it.
Hand off your project specification and let autonomous agents handle the rest — from planning to implementation.