Get Work Done: an agentic coding CLI for planning, execution, verification, and shipping.

GWD is a standalone CLI built on the Pi SDK. It manages project state, agent sessions, workflow tools, model routing, git isolation, recovery, and verification from one command-line entry point.

npm install -g @appfiex-rayliu/gwd@latest
gwd

GWD provisions a managed RTK binary on supported macOS, Linux, and Windows installs to compress shell-command output in bash, async_bash, bg_shell, and verification flows. GWD forces RTK_TELEMETRY_DISABLED=1 for all managed invocations. Set GWD_RTK_DISABLED=1 to disable the integration.

NOTICE: New to Node on Mac? If you installed Node.js via Homebrew, you may be running a development release instead of LTS. Read this guide to pin Node 24 LTS and avoid compatibility issues.

• Agentic project loop — plan milestones, execute tasks, verify results, and ship from one CLI.
• GWD runtime state — project artifacts live under .gwd/, with gwd.db as the authoritative runtime database and markdown projections for review.
• Current command surface — use gwd and /gwd commands only; old namespace aliases are intentionally not part of the initial release.
• Contributor-ready docs — user, GitBook, Mintlify, Docker, package, and developer docs now point to the rayliu-factory/gwd repository and current GWD package names.

Full documentation is in the docs/ directory:

• Getting Started — install, first run, basic usage
• Auto Mode — autonomous execution deep-dive
• Configuration — all preferences, models, git, and hooks
• Provider Setup — configure Ollama, vLLM Metal TurboQuant, LM Studio, and SGLang
• Custom Models — add custom providers (Ollama, vLLM, LM Studio, proxies)
• Token Optimization — profiles, context compression, complexity routing
• Cost Management — budgets, tracking, projections
• Git Strategy — worktree isolation, branching, merge behavior
• Parallel Orchestration — run multiple milestones simultaneously
• Working in Teams — unique IDs, shared artifacts
• Skills — bundled skills, discovery, custom authoring
• Commands Reference — all commands and keyboard shortcuts
• Troubleshooting — common issues, doctor, forensics, recovery
• Visualizer — workflow visualizer with stats and discussion status
• Remote Questions — route decisions to Slack or Discord when human input is needed
• Dynamic Model Routing — complexity-based model selection and budget pressure
• Web Interface — browser-based project management and real-time progress
• Import .planning Projects — import legacy .planning directories into .gwd
• Docker Sandbox — run GWD auto mode in an isolated Docker container

• Architecture — system design and dispatch pipeline
• Contributor Architecture and Data Flow — repo map, runtime flow, and agent-oriented change paths
• CI/CD Pipeline — three-stage promotion pipeline (Dev → Test → Prod)
• E2E Testing — real-process CLI/runtime coverage and CI runner expectations
• Pipeline Simplification (ADR-003) — merged research into planning, mechanical completion
• VS Code Extension — chat participant, sidebar dashboard, RPC integration

GWD structures work into a hierarchy:

Milestone  →  a shippable version (4-10 slices)
  Slice    →  one demoable vertical capability (1-7 tasks)
    Task   →  one context-window-sized unit of work

The iron rule: a task must fit in one context window. If it can't, it's two tasks.

Each slice flows through phases automatically:

Plan (with integrated research) → Execute (per task) → Complete → Reassess Roadmap → Next Slice
                                                                                      ↓ (all slices done)
                                                                              Validate Milestone → Complete Milestone

Plan scouts the codebase, researches relevant docs, and decomposes the slice into tasks with must-haves (mechanically verifiable outcomes). Execute runs each task in a fresh context window with only the relevant files pre-loaded — then runs configured verification commands (lint, test, etc.) with auto-fix retries. Complete writes the summary, UAT script, marks the roadmap, and commits with meaningful messages derived from task summaries. Reassess checks if the roadmap still makes sense given what was learned. Validate Milestone runs a reconciliation gate after all slices complete — comparing roadmap success criteria against actual results before sealing the milestone.

This is what makes GWD different. Run it, walk away, come back to built software.

/gwd auto

Auto mode is a state machine driven by the GWD database at the project root. It derives the next unit of work from authoritative SQLite state, creates a fresh agent session, injects a focused prompt with all relevant context pre-inlined, and lets the LLM execute. When the LLM finishes, auto mode persists the result to the database, refreshes markdown projections such as STATE.md, and dispatches the next unit.

The database is authoritative for milestones, slices, tasks, requirements, decisions, summaries, and completion status. Markdown under .gwd/ is a rendered projection for review, prompts, and git-friendly history; it is not a runtime fallback unless you explicitly run a recovery/import command. In worktree mode, project-root DB state remains authoritative and worktree markdown projections are not synced back as state.

What happens under the hood:

1. Fresh session per unit — Every task, every research phase, every planning step gets a clean 200k-token context window. No accumulated garbage. No "I'll be more concise now."

2. Context pre-loading — The dispatch prompt includes inlined task plans, slice plans, prior task summaries, dependency summaries, roadmap excerpts, and decisions register. The LLM starts with everything it needs instead of spending tool calls reading files.

3. Context Mode — Context Mode is enabled by default and gives eligible auto-mode units guidance for preserving context. Agents are steered toward gwd_exec for noisy scans, builds, tests, and diagnostics; capped stdout/stderr and metadata are saved under .gwd/exec/ while only a short digest enters the conversation. gwd_exec_search lets agents reuse prior runs instead of repeating expensive checks, and gwd_resume reads a prior compaction snapshot from .gwd/last-snapshot.md when one exists. Opt out with context_mode.enabled: false to disable Context Mode guidance, snapshot injection, gwd_exec, gwd_exec_search, and gwd_resume; tune sandbox timeout/output caps and environment forwarding with context_mode.exec_timeout_ms, context_mode.exec_stdout_cap_bytes, context_mode.exec_digest_chars, and context_mode.exec_env_allowlist.

4. Git isolation — When git.isolation is set to worktree or branch, each milestone runs on its own milestone/<MID> branch (in a worktree or in-place). All slice work commits sequentially — no branch switching, no merge conflicts. When the milestone completes, it's squash-merged to main as one clean commit. The default is none (work on the current branch), configurable via preferences. If worktree is configured in a repo with no committed HEAD, GWD temporarily behaves as none until the first commit exists because git worktrees need a committed start point.

5. Crash recovery — Auto mode persists worker state, unit-dispatch state, and paused-session metadata in the project-root SQLite database. If the session dies, the next /gwd auto reconstructs the interrupted unit from DB-backed runtime state, reads the surviving session file, synthesizes a recovery briefing from every tool call that made it to disk, and resumes with full context. Parallel orchestrator IPC still lives under .gwd/parallel/, so multi-worker sessions survive crashes too. In headless mode, crashes trigger automatic restart with exponential backoff (default 3 attempts).

6. Provider error recovery — Transient provider errors (rate limits, 500/503 server errors, overloaded) auto-resume after a delay. Permanent errors (auth, billing) pause for manual review. The model fallback chain retries transient network errors before switching models.

7. Stuck and artifact detection — A sliding-window detector identifies repeated dispatch patterns (including multi-unit cycles). Missing expected artifacts use a separate bounded path: GWD retries artifact verification up to 3 times with failure context, then pauses auto mode with the missing artifact error instead of looping indefinitely.

8. Timeout supervision — Soft timeout warns the LLM to wrap up. Idle watchdog detects stalls. Hard timeout pauses auto mode. Recovery steering nudges the LLM to finish durable output before giving up.

9. Cost tracking — Every unit's token usage and cost is captured, broken down by phase, slice, and model. The dashboard shows running totals and projections. Budget ceilings can pause auto mode before overspending.

10. Adaptive replanning — After each slice completes, the roadmap is reassessed. If the work revealed new information that changes the plan, slices are reordered, added, or removed before continuing.

11. Verification enforcement — Configure shell commands (npm run lint, npm run test, etc.) that run automatically after task execution. Failures trigger auto-fix retries before advancing. Auto-discovered checks from package.json run in advisory mode — they log warnings but don't block on pre-existing errors. Configurable via verification_commands, verification_auto_fix, and verification_max_retries preferences.

12. Milestone validation — After all slices complete, a validate-milestone gate compares roadmap success criteria against actual results before sealing the milestone.

13. Escape hatch — Press Escape to pause. The conversation is preserved. Interact with the agent, inspect what happened, or just /gwd auto to resume from disk state.

By default, /gwd runs in step mode: the same state machine as auto mode, but it pauses between units with a wizard showing what completed and what's next. You advance one step at a time, review the output, and continue when ready.

• No .gwd/ directory → Start a new project. Discussion flow captures your vision, constraints, and preferences.
• Milestone exists, no roadmap → Discuss or research the milestone.
• Roadmap exists, slices pending → Plan the next slice, execute one task, or switch to auto.
• Mid-task → Resume from where you left off.

/gwd next is an explicit alias for step mode. You can switch from step → auto mid-session via the wizard.

Step mode is the on-ramp. Auto mode is the highway.

npm install -g @appfiex-rayliu/gwd

First, choose your LLM provider:

gwd
/login

Select from 20+ providers — Anthropic, OpenAI, Google, OpenRouter, GitHub Copilot, and more. If you have a Claude Max or Copilot subscription, the OAuth flow handles everything. Otherwise, paste your API key when prompted.

GWD auto-selects a default model after login. To switch models later:

/model

Open a terminal in your project and run:

gwd

GWD opens an interactive agent session. From there, you have two ways to work:

/gwd — step mode. Type /gwd and GWD executes one unit of work at a time, pausing between each with a wizard showing what completed and what's next. Same state machine as auto mode, but you stay in the loop. No project yet? It starts the discussion flow. Roadmap exists? It plans or executes the next step.

/gwd auto — autonomous mode. Type /gwd auto and walk away. GWD researches, plans, executes, verifies, commits, and advances through every slice until the milestone is complete. Fresh context window per task. No babysitting.

The real workflow: run auto mode in one terminal, steer from another.

Terminal 1 — let it build

gwd
/gwd auto

Terminal 2 — steer while it works

gwd
/gwd discuss    # talk through architecture decisions
/gwd status     # check progress
/gwd queue      # queue the next milestone

Both terminals coordinate through the same project-root GWD runtime on local disk. The SQLite database is authoritative, .gwd/ markdown is refreshed from it, and your decisions in terminal 2 are picked up at the next phase boundary without stopping auto mode.

gwd headless runs any /gwd command without a TUI. Designed for CI pipelines, cron jobs, and scripted automation.

# Run auto mode in CI
gwd headless --timeout 600000

# Create and execute a milestone end-to-end
gwd headless new-milestone --context spec.md --auto

# One unit at a time (cron-friendly)
gwd headless next

# Instant JSON snapshot (no LLM, ~50ms)
gwd headless query

# Force a specific pipeline phase
gwd headless dispatch plan

Headless auto-responds to interactive prompts, detects completion, and exits with structured codes: 0 complete, 1 error/timeout, 2 blocked. Auto-restarts on crash with exponential backoff. Use gwd headless query for instant, machine-readable state inspection — returns phase, next dispatch preview, and parallel worker costs as a single JSON object without spawning an LLM session. Pair with remote questions to route decisions to Slack or Discord when human input is needed.

Multi-session orchestration — headless mode supports DB-backed coordination across multiple GWD workers on the same machine. Worker registration, milestone leases, unit dispatch tracking, and command delivery live in .gwd/gwd.db, while .gwd/parallel/ remains a local runtime area for per-milestone locks and isolation artifacts.

On first run, GWD launches a branded setup wizard that walks you through LLM provider selection (OAuth or API key), then optional tool API keys (Brave Search, Context7, Jina, Slack, Discord). Every step is skippable — press Enter to skip any. If you have an existing Pi installation, your provider credentials (LLM and tool keys) are imported automatically. Run gwd config anytime to re-run the wizard.

Every dispatch is carefully constructed. The LLM never wastes tool calls on orientation.

Branch-per-slice with squash merge. Fully automated.

main:
  docs(M001/S04): workflow documentation and examples
  fix(M001/S03): bug fixes and doc corrections
  feat(M001/S02): API endpoints and middleware
  feat(M001/S01): data model and type system

gwd/M001/S01 (deleted after merge):
  feat(S01/T03): file writer with round-trip fidelity
  feat(S01/T02): markdown parser for plan files
  feat(S01/T01): core types and interfaces

One squash commit per milestone on main (or whichever branch you started from). The worktree is torn down after merge. Git bisect works. Individual milestones are revertable. Commit messages are generated from task summaries — no more generic "complete task" messages.

Every task has must-haves — mechanically checkable outcomes:

• Truths — Observable behaviors ("User can sign up with email")
• Artifacts — Files that must exist with real implementation, not stubs
• Key Links — Imports and wiring between artifacts

The verification ladder: static checks → command execution → behavioral testing → human review (only when the agent genuinely can't verify itself).

Ctrl+Alt+G or /gwd status opens a real-time overlay showing:

• Current milestone, slice, and task progress
• Auto mode elapsed time and phase
• Per-unit cost and token breakdown by phase, slice, and model
• Cost projections based on completed work
• Completed and in-progress units

After a milestone completes, GWD auto-generates a self-contained HTML report in .gwd/reports/. Each report includes project summary, progress tree, slice dependency graph (SVG DAG), cost/token metrics with bar charts, execution timeline, changelog, and knowledge base sections. No external dependencies — all CSS and JS are inlined, printable to PDF from any browser.

An auto-generated index.html shows all reports with progression metrics across milestones.

• Automatic — generated after milestone completion (configurable via auto_report preference)
• Manual — run /gwd export --html anytime

GWD preferences live in ~/.gwd/PREFERENCES.md (global) or .gwd/PREFERENCES.md (project). Manage with /gwd prefs.

---
version: 1
models:
  research: claude-sonnet-4-6
  planning:
    model: claude-opus-4-7
    fallbacks:
      - openrouter/z-ai/glm-5
      - openrouter/minimax/minimax-m2.5
  execution: claude-sonnet-4-6
  completion: claude-sonnet-4-6
skill_discovery: suggest
auto_supervisor:
  soft_timeout_minutes: 20
  idle_timeout_minutes: 10
  hard_timeout_minutes: 30
budget_ceiling: 50.00
unique_milestone_ids: true
verification_commands:
  - npm run lint
  - npm run test
auto_report: true
---

Key settings:

Place an AGENTS.md file in any directory to provide persistent behavioral guidance for that scope. Pi core loads AGENTS.md automatically (with CLAUDE.md as a fallback) at both user and project levels. Use these files for coding standards, architectural decisions, domain terminology, or workflow preferences.

Note: The legacy agent-instructions.md format (~/.gwd/agent-instructions.md and .gwd/agent-instructions.md) is deprecated and no longer loaded. Migrate any existing instructions to AGENTS.md or CLAUDE.md.

Start GWD with gwd --debug to enable structured JSONL diagnostic logging. Debug logs capture dispatch decisions, state transitions, and timing data for troubleshooting auto-mode issues.

GWD includes a coordinated token optimization system that reduces usage by 40-60% on cost-sensitive workloads. Set a single preference to coordinate model tier selection, phase skipping, and context compression:

token_profile: budget # or balanced (default), quality

Complexity-based routing automatically classifies tasks as simple/standard/complex and routes to appropriate available models. Token profiles define provider-agnostic tier intentions, so simple docs tasks use a light-tier configured model and complex architectural work can use a heavy-tier configured model. The classification is heuristic (sub-millisecond, no LLM calls) and learns from outcomes via a persistent routing history.

Budget pressure graduates model downgrading as you approach your budget ceiling — 50%, 75%, and 90% thresholds progressively shift work to cheaper tiers.

See the full Token Optimization Guide for details.

GWD ships with 24 extensions, all loaded automatically:

Five specialized subagents for delegated work:

The best practice for working in teams is to ensure unique milestone names across all branches (by using unique_milestone_ids) and checking in the right .gwd/ artifacts to share valuable context between teammates.

# ── GWD: Runtime / Ephemeral (per-developer, per-session) ──────────────────
# Auto-mode dispatch tracker — prevents re-running completed units (includes archived per-milestone files)
.gwd/completed-units*.json
# State manifest — workflow state for recovery
.gwd/state-manifest.json
# Derived state projection — regenerated from the authoritative database
.gwd/STATE.md
# Per-developer token/cost accumulator
.gwd/metrics.json
# Raw JSONL session dumps — crash recovery forensics, auto-pruned
.gwd/activity/
# Unit execution records — dispatch phase, timeouts, and recovery tracking
.gwd/runtime/
# Git worktree working copies
.gwd/worktrees/
# Parallel runtime locks and per-milestone isolation artifacts
.gwd/parallel/
# SQLite database and WAL sidecars — authoritative runtime state, local only
.gwd/gwd.db*
# Daily-rotated event journal — structured event log for forensics
.gwd/journal/
# Doctor run history — diagnostic check results
.gwd/doctor-history.jsonl
# Workflow event log — structured event stream
.gwd/event-log.jsonl
# Generated HTML reports (regenerable via /gwd export --html)
.gwd/reports/
# Session-specific interrupted-work markers
.gwd/milestones/**/continue.md
.gwd/milestones/**/*-CONTINUE.md

Create or amend your .gwd/PREFERENCES.md file within the repo to include unique_milestone_ids: true e.g.

---
version: 1
unique_milestone_ids: true
---

With the above .gitignore set up, the .gwd/PREFERENCES.md file is checked into the repo ensuring all teammates use unique milestone names to avoid collisions.

Milestone names will now be generated with a 6 char random string appended e.g. instead of M001 you'll get something like M001-ush8s3

1. Ensure you are not in the middle of any milestones (clean state)
2. Update the .gwd/ related entries in your .gitignore to follow the Suggested .gitignore setup section under Working in teams (ensure you are no longer blanket ignoring the whole .gwd/ directory)
3. Update your .gwd/PREFERENCES.md file within the repo as per section Unique Milestone Names
4. If you want to update all your existing milestones use this prompt in GWD: I have turned on unique milestone ids, please update all old milestone ids to use this new format e.g. M001-abc123 where abc123 is a random 6 char lowercase alpha numeric string. Update all references in all .gwd file contents, file names and directory names. Validate your work once done to ensure referential integrity.
5. Commit to git

GWD is a TypeScript application that embeds the Pi coding agent SDK.

gwd (CLI binary)
  └─ loader.ts          Sets PI_PACKAGE_DIR, GWD env vars, dynamic-imports cli.ts
      └─ cli.ts         Wires SDK managers, loads extensions, starts InteractiveMode
          ├─ headless.ts     Headless orchestrator (spawns RPC child, auto-responds, detects completion)
          ├─ onboarding.ts   First-run setup wizard (LLM provider + tool keys)
          ├─ wizard.ts       Env hydration from stored auth.json credentials
          ├─ app-paths.ts    ~/.gwd/agent/, ~/.gwd/sessions/, auth.json
          ├─ resource-loader.ts  Syncs bundled extensions + agents to ~/.gwd/agent/
          └─ src/resources/
              ├─ extensions/gwd/    Core GWD extension (auto, state, commands, ...)
              ├─ extensions/...     21 supporting extensions
              ├─ agents/            scout, researcher, worker, javascript-pro, typescript-pro
              └─ GWD-WORKFLOW.md    Manual bootstrap protocol

Key design decisions:

• pkg/ shim directory — PI_PACKAGE_DIR points here (not project root) to avoid Pi's theme resolution collision with our src/ directory. Contains only piConfig and theme assets.
• Two-file loader pattern — loader.ts sets all env vars with zero SDK imports, then dynamic-imports cli.ts which does static SDK imports. This ensures PI_PACKAGE_DIR is set before any SDK code evaluates.
• Always-overwrite sync — npm update -g takes effect immediately. Bundled extensions and agents are synced to ~/.gwd/agent/ on every launch, not just first run.
• DB-authoritative state — the project-root GWD database is the runtime source of truth. .gwd/ markdown files are rendered projections for review, prompt context, and git history. No in-memory state survives across sessions.

• Node.js ≥ 22.0.0 (24 LTS recommended)
• An LLM provider — any of the 20+ supported providers (see Use Any Model)
• Git — initialized automatically if missing

Optional:

• Brave Search API key (web research)
• Tavily API key (web research — alternative to Brave)
• Google Gemini API key (web research via Gemini Search grounding)
• Context7 API key (library docs)
• Jina API key (page extraction)

GWD isn't locked to one provider. It runs on the Pi SDK, which supports 20+ model providers out of the box. Use different models for different phases — Opus for planning, Sonnet for execution, a fast model for research.

Anthropic, Anthropic (Vertex AI), OpenAI, Google (Gemini), OpenRouter, GitHub Copilot, Amazon Bedrock, Azure OpenAI, Google Vertex, Groq, Cerebras, Mistral, xAI, HuggingFace, Vercel AI Gateway, and more.

If you have a Claude Max, Codex, or GitHub Copilot subscription, you can use those directly — Pi handles the OAuth flow. No API key needed.

⚠️ Important: Using OAuth tokens from subscription plans outside their native applications may violate the provider's Terms of Service. In particular:

• Google Gemini — Using Gemini CLI or Antigravity OAuth tokens in third-party tools has resulted in Google account suspensions. This affects your entire Google account, not just the Gemini service. Use a Gemini API key instead.
• Claude Max — Anthropic's ToS may not explicitly permit OAuth use outside Claude's own applications.
• GitHub Copilot — Usage outside GitHub's own tools may be restricted by your subscription terms.

GWD supports API key authentication for all providers as the safe alternative. We strongly recommend using API keys over OAuth for Google Gemini.

OpenRouter gives you access to hundreds of models through a single API key. Use it to run GWD with Llama, DeepSeek, Qwen, or anything else OpenRouter supports.

In your preferences (/gwd prefs), assign different models to different phases:

models:
  research: openrouter/deepseek/deepseek-r1
  planning:
    model: claude-opus-4-7
    fallbacks:
      - openrouter/z-ai/glm-5
  execution: claude-sonnet-4-6
  completion: claude-sonnet-4-6

Use expensive models where quality matters (planning, complex execution) and cheaper/faster models where speed matters (research, simple completions). Each phase accepts a simple model string or an object with model and fallbacks — if the primary model fails (provider outage, rate limit, credit exhaustion), GWD automatically tries the next fallback. GWD tracks cost per-model so you can see exactly where your budget goes.

MIT License