Boatman Ecosystem documentation is live!
BoatmanMode CLI
Overview

BoatmanMode CLI

BoatmanMode is an AI-powered development agent that automates ticket execution with peer review. It fetches tasks from Linear (or accepts inline prompts and files), generates code using Claude, reviews changes with a configurable Claude skill, iterates until quality passes, and creates pull requests.

Key Features

AI-Powered Development

  • Generates complete implementations from ticket descriptions or prompts
  • Understands project conventions via Claude's context window
  • Creates appropriate tests alongside production code

Peer Review with Claude Skill

  • Uses a configurable Claude skill for code review (default: peer-review)
  • Automated pass/fail verdict with detailed feedback
  • Customize via --review-skill flag or config

Iterative Refinement

  • Automatically refactors based on review feedback
  • Fresh agent per iteration — clean context, no token bloat
  • Structured handoffs between agents with concise context

Live Activity Streaming

  • Watch Claude work in real-time via tmux sessions
  • See every tool call: file reads, edits, bash commands
  • Full visibility into AI decision-making

Git Worktree Isolation

  • Each task works in an isolated git worktree
  • No interference with your main working directory
  • Safe for parallel task execution

Agent Pipeline Overview

The workflow orchestrates multiple AI agents through a structured pipeline:

Step 1: PLANNER AGENT      → Analyzes codebase, creates plan
Step 2: PREFLIGHT VALIDATION → Validates plan before execution
Step 3: EXECUTOR AGENT      → Implements code changes
Step 4: TEST RUNNER          → Detects framework, runs tests
Step 5: REVIEWER AGENT      → Reviews diff, pass/fail verdict
Step 6: REFACTOR AGENT      → Fixes issues (loops until pass)
Step 7: DIFF VERIFICATION   → Verifies issues were addressed
Step 8: COMMIT & PUSH       → Commits and pushes changes
Step 9: CREATE PR            → Opens pull request

Each step saves a checkpoint for crash recovery. Patterns are learned on success for cross-session improvement.


Support Systems

SystemPurpose
Context PinningEnsures consistency during multi-file changes
Dynamic Handoff CompressionAdapts context size to token budgets
Smart File SummarizationHandles large files intelligently
Issue DeduplicationTracks issues across review iterations
Git-Integrated CheckpointsSaves progress as git commits
Agent MemoryCross-session learning for improved performance

Project Structure

cli/
├── cmd/boatman/main.go       # Entry point
├── internal/
│   ├── agent/                # Workflow orchestration
│   ├── checkpoint/           # Progress saving/resume
│   ├── claude/               # Claude CLI wrapper
│   ├── cli/                  # Cobra commands
│   ├── config/               # Configuration
│   ├── contextpin/           # File dependency tracking
│   ├── coordinator/          # Parallel agent coordination
│   ├── diffverify/           # Diff verification agent
│   ├── events/               # JSON event emission
│   ├── executor/             # Code generation
│   ├── filesummary/          # Smart file summarization
│   ├── github/               # PR creation (gh CLI)
│   ├── handoff/              # Agent context passing
│   ├── healthcheck/          # Dependency verification
│   ├── issuetracker/         # Issue deduplication
│   ├── linear/               # Linear API client
│   ├── logger/               # Structured logging
│   ├── memory/               # Cross-session learning
│   ├── planner/              # Plan generation
│   ├── preflight/            # Pre-execution validation
│   ├── retry/                # Exponential backoff
│   ├── scottbott/            # Peer review
│   ├── testenv/              # E2E test environment
│   ├── testrunner/           # Test execution
│   ├── tmux/                 # Session management
│   └── worktree/             # Git worktree management
└── examples/                 # Library usage examples