Apex Forge

An execution framework for AI coding agents. 43 skills, three complexity tiers, one install. Remixed from 8 open-source projects.

Why

AI coding agents are unreliable. They guess instead of verifying. They skip tests. They claim "done" without proof. They can't coordinate on multi-step tasks. And you end up installing 5 different tools to get a complete workflow.

Apex Forge fixes this by enforcing discipline at the protocol level, not by hoping the agent behaves.

Quick Start

git clone https://github.com/d-wwei/apex-forge.git ~/.claude/skills/apex-forge
cd ~/.claude/skills/apex-forge
bun install && bun run build:all
bunx playwright install chromium
./dist/apex-forge init

Or just tell the AI: "initialize apex forge"

You now have 3 binaries (apex-forge, apex-forge-browse, apex-forge-mcp), 43 skills, and a protocol that auto-injects on every session.

After init, your project gets an .apex/ directory:

.apex/
  state.json      -- current stage and session state
  tasks.json      -- task list and state machine
  memory.json     -- memory store (persists across sessions)
  analytics/      -- usage data
  screenshots/    -- browser screenshots

Already in .gitignore. Won't be committed.

Usage

Every scenario below shows two ways to work: talk to the AI in natural language, or run a command directly. Mix and match.

Build a Feature

Tell the AI: "I want to add user auth with JWT"

Or drive each stage manually:

# 1. Requirements exploration (hard gate: no code gets written here)
/apex-forge-brainstorm
# -> produces docs/brainstorms/auth-requirements.md

# 2. Implementation plan (file paths, function signatures, test scenarios)
/apex-forge-plan
# -> produces docs/plans/auth-plan.md

# 3. TDD execution (tests first, complex tasks auto-split to sub-agents)
/apex-forge-execute

# 4. Quality review (18 perspectives: security, correctness, framework-specific)
/apex-forge-review

# 5. Ship (run tests -> version bump -> CHANGELOG -> commit -> PR)
/apex-forge-ship

# 6. Knowledge capture (written to docs/solutions/ for future reuse)
/apex-forge-compound

You don't need to remember the order. After each stage, the AI suggests the next one.

You say	Stage triggered	Command
"Help me explore the requirements"	Requirements	`/apex-forge-brainstorm`
"Create an implementation plan"	Planning	`/apex-forge-plan`
"Start coding"	Execution	`/apex-forge-execute`
"Review my changes"	Review	`/apex-forge-review`
"Ready to ship"	Ship	`/apex-forge-ship`
"Save what we learned"	Knowledge capture	`/apex-forge-compound`

Fix a Bug

Tell the AI: "Login returns 401 but the token is valid"

Or run: /apex-forge-investigate

The investigation follows a fixed sequence: reproduce the issue, add logging at code boundaries, form 3 hypotheses and test each, confirm root cause, then fix with a regression test. Rule: no fix without root cause.

Code Review

Tell the AI: "Review my changes"

Or run: /apex-forge-code-review

Reads git diff, checks from 18 perspectives. Changes to .tsx files activate the React reviewer. Changes to .py activate Python. Verdict: SHIP / SHIP_WITH_FIXES / BLOCK.

QA Testing

Tell the AI: "Test this page"

Or run: /apex-forge-qa

Three depths: Quick (critical issues only), Standard (+medium), Exhaustive (full coverage). If a browser is available, it opens the page, takes screenshots, interacts, and verifies.

Security Audit

Tell the AI: "Run a security check"

Or run: /apex-forge-security-audit

Checks in order: leaked secrets, dependency vulnerabilities, CI/CD security, OWASP Top 10, auth/authz.

More Scenarios

You say	Command	What happens
"Is this plan ambitious enough?"	`/apex-forge-ceo-review`	CEO/founder perspective on scope
"Is the architecture right?"	`/apex-forge-eng-review`	Engineering review: data, API, performance, deployment
"Does the design look good?"	`/apex-forge-design-review`	Visual QA with screenshot comparison
"Review this week"	`/apex-forge-retro`	Git stats + team retrospective
"Open the dashboard"	`apex-forge dashboard`	Web Kanban + activity stream + analytics
"Run this code safely"	`apex-forge sandbox js "..."`	Docker-sandboxed execution
"Import GitHub issues"	`apex-forge issues import`	Import as Apex tasks
"Create a new skill"	`/apex-forge-skill-author`	Guided skill authoring
"Show me 3 design directions"	`/apex-forge-design-shotgun`	3 visual approaches for one requirement
"Full plan review"	`/apex-forge-autoplan`	Auto-run CEO + engineering + design reviews

Complete CLI reference: Usage Guide

Task Management

Tell the AI "create a task: implement user auth", or use the CLI:

# Create
apex-forge task create "Implement user auth" "JWT middleware + refresh tokens"
apex-forge task create "Write auth tests" "Integration tests" T1    # T1 is a dependency

# View
apex-forge task list                        # all tasks
apex-forge task list --status open          # only open
apex-forge task next                        # next unblocked task
apex-forge task get T1                      # details

# State transitions (enforced -- no skipping steps)
apex-forge task assign T1                   # open -> assigned
apex-forge task start T1                    # assigned -> in_progress
apex-forge task submit T1 "Tests passing"   # in_progress -> to_verify
apex-forge task verify T1 pass              # to_verify -> done
apex-forge task verify T1 fail              # to_verify -> in_progress (redo)
apex-forge task block T1 "Waiting for API key"
apex-forge task release T1                  # assigned -> open (unclaim)

Memory System

Tell the AI "remember: auth uses JWT RS256", or use the CLI:

# Store (confidence 0.0-1.0 + tags)
apex-forge memory add "Auth uses JWT RS256" 0.9 auth jwt
apex-forge memory add "Database is PostgreSQL 16" 0.85 db

# Query
apex-forge memory list                      # all facts
apex-forge memory list --min 0.8            # high-confidence only
apex-forge memory search "JWT"              # keyword search

# Maintain
apex-forge memory curate                    # auto-extract from git/tasks/solutions
apex-forge memory prune                     # clean low-quality (<0.5) entries

# In-session curation (AI reviews the current session)
/apex-forge-memory-curate

High-confidence facts auto-inject into session context. Low-confidence entries get pruned automatically.

Browser

Tell the AI "open https://my-app.com and check it", or use the CLI:

# Navigate
apex-forge-browse goto https://your-app.com
apex-forge-browse text                      # read page text
apex-forge-browse links                     # list all links

# Interact (snapshot first to see elements, then use @e refs)
apex-forge-browse snapshot -i               # list interactive elements -> @e1, @e2, ...
apex-forge-browse click @e3                 # click
apex-forge-browse fill @e5 "test@test.com"  # fill input

# Screenshot
apex-forge-browse screenshot /tmp/page.png
apex-forge-browse responsive /tmp/layout    # auto-capture mobile/tablet/desktop

# Inspect
apex-forge-browse console --errors          # JS errors
apex-forge-browse is visible ".modal"       # element visibility check
apex-forge-browse perf                      # page load performance

# Cookies (for testing authenticated pages)
apex-forge-browse cookie-import-browser     # import from real Chrome

# Visible mode (real Chrome window + side panel showing AI actions)
apex-forge-browse connect
apex-forge-browse disconnect                # switch back to headless

Dashboard

Tell the AI "open the dashboard", or:

apex-forge dashboard                        # default port 3456
apex-forge dashboard --port 8080            # custom port

Five panels: task Kanban (5-column drag-and-drop), pipeline state (current stage + artifacts), activity stream (real-time skill execution), memory panel (facts sorted by confidence), and analytics (skill usage, average time, success rate).

All 43 Skills

Protocol (3)

Skill	What it does
`/apex-forge`	Core execution protocol. Complexity router, TDD enforcement, evidence grading, escalation ladder, verification gate. Auto-activates every session.
`/apex-forge-round`	PDCA round-based execution for multi-step tasks. Named round types: clarify, explore, hypothesis, planning, execution, verification, hardening, recovery.
`/apex-forge-wave`	Wave-based delivery for project-scale work. Cross-session state, assumption registry, decision log, handoff protocol.

Stages (7)

Skill	What it does
`/apex-forge-brainstorm`	Requirements exploration. Hard gate: no code until design is approved.
`/apex-forge-plan`	Implementation plan with file paths, tasks, test scenarios. Checks for brainstorm artifact.
`/apex-forge-execute`	TDD-first implementation. Sub-agent dispatch for complex tasks. Checks for plan artifact.
`/apex-forge-review`	3-persona quality gate. Checks git diff.
`/apex-forge-ship`	Tests, version bump, changelog, commit, PR.
`/apex-forge-compound`	Knowledge capture to `docs/solutions/` with overlap detection.
`/apex-forge-verify`	5-step evidence gate. Standalone -- usable anytime.

Quality and Review (6)

Skill	What it does
`/apex-forge-qa`	Systematic QA testing with tiered depth and browser-aware verification.
`/apex-forge-investigate`	Root-cause debugging. No fixes without understanding why.
`/apex-forge-code-review`	Multi-pass code review: correctness, security, performance, maintainability.
`/apex-forge-design-review`	Visual QA. Screenshot-driven fix-verify loops.
`/apex-forge-security-audit`	Infrastructure-first security audit.
`/apex-forge-retro`	Retrospective analysis. Extract lessons, patterns, improvements.

Plan Review (4)

Skill	What it does
`/apex-forge-ceo-review`	Strategic plan review from a CEO/founder perspective.
`/apex-forge-eng-review`	Engineering architecture review.
`/apex-forge-plan-design-review`	Designer's eye plan review with dimension ratings.
`/apex-forge-autoplan`	Runs CEO, Eng, and Design reviews sequentially with auto-decisions.

Creative (3)

Skill	What it does
`/apex-forge-design-consultation`	Design system creation: aesthetic direction, tokens, typography, color, preview.
`/apex-forge-design-shotgun`	Generate 3 distinct visual approaches for a UI requirement.
`/apex-forge-office-hours`	Guided learning. Teach concepts in the context of the codebase.

Operations (5)

Skill	What it does
`/apex-forge-canary`	Post-deploy canary monitoring. Screenshots, error detection, performance regression checks.
`/apex-forge-benchmark`	Performance baseline tracking. Measure, store, compare, flag regressions.
`/apex-forge-land-and-deploy`	Merge PR, wait for CI, deploy, verify with canary checks.
`/apex-forge-setup-deploy`	Auto-detect and configure deployment: platform, URLs, health checks.
`/apex-forge-document-release`	Post-ship documentation sync.

Safety (4)

Skill	What it does
`/apex-forge-guard`	Freeze edits to a directory + destructive command warnings.
`/apex-forge-freeze`	Restrict edits to a specific directory. Writes outside the boundary are refused.
`/apex-forge-unfreeze`	Remove the freeze boundary.
`/apex-forge-careful`	Require confirmation before destructive operations.

Browser (3)

Skill	What it does
`/apex-forge-browse`	Browser interaction: navigate, read, interact, screenshot.
`/apex-forge-connect-chrome`	Launch Chrome with remote debugging for automation.
`/apex-forge-setup-browser-cookies`	Import cookies from a real browser for authenticated headless sessions.

Knowledge (3)

Skill	What it does
`/apex-forge-memory-curate`	Review, score, and prune the memory store.
`/apex-forge-compound-refresh`	Update stale solution docs.
`/apex-forge-skill-author`	Guided skill file creation.

Orchestration (3)

Skill	What it does
`/apex-forge-wave-planner`	Plan wave-based delivery with task decomposition.
`/apex-forge-wave-challenger`	Challenge assumptions in wave plans.
`/apex-forge-wave-worker`	Execute assigned wave tasks.

External (2)

Skill	What it does
`/apex-forge-codex-consult`	Second opinion via OpenAI Codex CLI or independent subagent.
`/apex-forge-mobile-test`	Mobile device testing.

CLI Reference

Key commands. For the full list, run apex-forge help.

# Project
apex-forge init                              # initialize .apex/ in current project
apex-forge status                            # pipeline state, active tasks, memory stats
apex-forge recover                           # fix stale state after crashes

# Tasks
apex-forge task create TITLE [DESC] [DEPS]   # create a task
apex-forge task list [--status STATUS]        # list tasks
apex-forge task next                         # next unblocked task
apex-forge task assign|start|submit|verify|block|release ID

# Memory
apex-forge memory add CONTENT SCORE TAGS     # store a fact
apex-forge memory list [--min SCORE]         # list facts
apex-forge memory search QUERY               # keyword search
apex-forge memory curate                     # auto-extract knowledge
apex-forge memory prune                      # clean low-quality entries

# Browser
apex-forge-browse goto URL                   # navigate
apex-forge-browse snapshot -i                # list interactive elements
apex-forge-browse click|fill|screenshot      # interact
apex-forge-browse connect|disconnect         # visible/headless mode toggle

# Dashboard
apex-forge dashboard [--port PORT]           # web UI

# Orchestration
apex-forge orchestrate [--dry-run] [--once]  # multi-agent task dispatch

# Other
apex-forge telemetry report                  # skill usage stats
apex-forge worktree create|list|cleanup      # git worktree management
apex-forge sandbox js|python|bash CODE       # sandboxed execution
apex-forge issues list|import|view           # GitHub issue sync
apex-forge convert --platform PLATFORM       # export to Cursor/Codex/Gemini/Windsurf/Factory
apex-forge consensus test-all                # run all consensus protocol tests

Complete CLI reference: Usage Guide

Core Concepts

Complexity Router

Not every task needs the full ceremony:

Tier	When	Flow
Tier 1 (single pass)	Simple fixes	Do it, verify, done
Tier 2 (rounds)	Multi-step work	PDCA rounds with named types, max 5 rounds
Tier 3 (waves)	Project-scale	Cross-session waves, persistent state in `.apex/state.json`

The AI auto-detects which tier fits. You can override with default_tier in config.

Escalation Ladder

Failure automatically raises the strictness level:

Level	Trigger	Requirement
L0	Normal	Standard protocol
L1	2nd failure	Fundamentally different approach
L2	3rd failure	3 testable hypotheses
L3	4th failure	7-point recovery checklist
L4	5th failure	Minimal reproduction, hand off to human

Evidence Grading

Grade	Meaning	Minimum for
E0	Guess	Hypothesis only
E1	Indirect evidence	Hypothesis only
E2	Direct evidence	Taking action
E3	Multi-source verified	Claiming success
E4	Verified + reproduced	Highest confidence

Verification Gate

Five steps before claiming "done":

Identify what command would prove the claim
Run it now (not old results)
Read the full output
Does the output confirm the claim? Yes or no
Only "yes" counts as done

Skip any step and it's not verification, it's guessing.

TDD Iron Rule

No production code without a failing test. Write test, RED (confirm it fails for the right reason), GREEN, refactor. 14 rationalizations are pre-blocked.

Knowledge Compounding

Every solved problem gets written to docs/solutions/. Overlap detection prevents duplicates. Stale docs get auto-refreshed. Over time, the system gets cheaper to run.

Installation

Prerequisites: Bun 1.3+

Claude Code (global -- recommended)

git clone https://github.com/d-wwei/apex-forge.git ~/.claude/skills/apex-forge
cd ~/.claude/skills/apex-forge
bun install && bun run build:all
bunx playwright install chromium

The protocol auto-activates on every session via the hooks/session-start hook.

Claude Code (project-local)

git clone https://github.com/d-wwei/apex-forge.git .claude/skills/apex-forge
cd .claude/skills/apex-forge
bun install && bun run build:all
echo ".claude/skills/apex-forge" >> .gitignore

Cursor

git clone https://github.com/d-wwei/apex-forge.git .cursor-plugin/apex-forge
cd .cursor-plugin/apex-forge && bun install && bun run build:all

Codex / Factory

git clone https://github.com/d-wwei/apex-forge.git .agents/skills/apex-forge
cd .agents/skills/apex-forge && bun install && bun run build:all

Gemini / Windsurf

git clone https://github.com/d-wwei/apex-forge.git apex-forge
cd apex-forge && bun install && bun run build:all
./dist/apex-forge convert --platform gemini   # or --platform windsurf

Auto-detect

./setup    # Detects installed platforms, creates symlinks, writes .apex/config.yaml

Configuration

Create .apex/config.yaml in your project root (auto-created by ./setup or apex-forge init):

default_tier: auto          # auto | 1 | 2 | 3
proactive: true             # suggest next stages automatically
compound_on_resolve: true   # auto-trigger knowledge capture after ship
max_concurrent_agents: 3    # for parallel sub-agent dispatch
autonomy: balanced          # high | balanced | controlled
solutions_dir: docs/solutions/

Architecture

apex-forge/
  src/                        TypeScript, compiled to 3 binaries
    cli.ts                    Main CLI (apex-forge)
    commands/                 Task, memory, status, telemetry, worktree handlers
    browse/                   Headless browser daemon (apex-forge-browse)
    mcp/                      MCP server with role-based tools (apex-forge-mcp)
    consensus/                Raft, BFT, Gossip, CRDT implementations
    integrations/             GitHub issue tracker
    dashboard.ts              Web dashboard server
    orchestrator.ts           Multi-agent task dispatcher
    sandbox.ts                Sandboxed code execution (JS/Python/Bash)
    converter.ts              Cross-platform config export
    tracing.ts                Observability spans
    design.ts                 AI design generation

  workflow/                   43 skill files (Markdown with YAML frontmatter)
    stages/                   7 pipeline stages
    roles/                    34 role-based skills

  protocol/                   Core execution discipline
    SKILL.md                  Auto-injected protocol (14 sections)
    round-based-execution.md  PDCA rounds for Tier 2
    wave-based-delivery.md    Wave delivery for Tier 3

  orchestration/              Multi-agent coordination
    ARCHITECTURE.md           Architecture spec (10 sections)
    PATTERNS.md               10 reusable orchestration patterns
    registry-seeds.yaml       115 agent templates

  extension/                  Chrome Side Panel
    manifest.json             Extension manifest
    sidepanel.html/js         Side Panel UI
    background.js             Service worker
    content.js/css            Content script

  hooks/                      Session and git hooks
    session-start             Auto-injects protocol on every session
    state-helper              Bash functions for state, memory, worktree, telemetry
    task-helper               Task state machine operations

  platforms/                  Platform-specific manifests
    cursor/                   Cursor plugin
    codex/                    Codex agent config
    factory/                  Factory droid config

Provenance

Every pattern traces to a specific source project.

Capability	Source	License
Auto-triggering, TDD enforcement, verification gate	superpowers	MIT
Complexity router, escalation ladder, evidence grading	better-work-skill	MIT
Phase discipline, knowledge compounding, multi-persona review	compound-engineering	MIT
Role-based skills, telemetry, proactive routing	gstack	MIT
Single-authority orchestrator, workspace isolation	symphony	MIT
Task state machine, role enforcement, dual-stage acceptance	chorus	MIT
LLM-curated memory, sandbox propagation, subagent safety	deer-flow	Apache 2.0
Ledger pattern, skills registry, cost-aware routing	ruflo	MIT

Full attribution: docs/PROVENANCE.md

Troubleshooting

Commands don't appear?

ls -la ~/.claude/skills/apex-forge          # check symlink
ln -sf /path/to/apex-forge ~/.claude/skills/apex-forge  # re-link

Browser errors?

cd ~/.claude/skills/apex-forge && bunx playwright install chromium

Dashboard won't open?

apex-forge dashboard --port 3456 && open http://localhost:3456

State corruption?

apex-forge recover                          # auto-fix

Stats

Metric	Count
Source files (excluding deps)	128
Lines of code	~28,000
Automated tests	55
Skill files	43
Agent templates	115
Browser commands	59
Consensus protocols	4

License

MIT -- see LICENSE.

Name		Name	Last commit message	Last commit date
Latest commit History 66 Commits
.github/workflows		.github/workflows
apex-forge-autoplan		apex-forge-autoplan
apex-forge-benchmark		apex-forge-benchmark
apex-forge-brainstorm		apex-forge-brainstorm
apex-forge-browse		apex-forge-browse
apex-forge-canary		apex-forge-canary
apex-forge-careful		apex-forge-careful
apex-forge-code-review		apex-forge-code-review
apex-forge-codex-consult		apex-forge-codex-consult
apex-forge-compound-refresh		apex-forge-compound-refresh
apex-forge-compound		apex-forge-compound
apex-forge-connect-chrome		apex-forge-connect-chrome
apex-forge-design-consultation		apex-forge-design-consultation
apex-forge-design-review		apex-forge-design-review
apex-forge-design-shotgun		apex-forge-design-shotgun
apex-forge-design-to-code		apex-forge-design-to-code
apex-forge-document-release		apex-forge-document-release
apex-forge-execute		apex-forge-execute
apex-forge-freeze		apex-forge-freeze
apex-forge-guard		apex-forge-guard
apex-forge-investigate		apex-forge-investigate
apex-forge-land-and-deploy		apex-forge-land-and-deploy
apex-forge-mobile-test		apex-forge-mobile-test
apex-forge-office-hours		apex-forge-office-hours
apex-forge-plan-ceo-review		apex-forge-plan-ceo-review
apex-forge-plan-design-review		apex-forge-plan-design-review
apex-forge-plan-eng-review		apex-forge-plan-eng-review
apex-forge-plan		apex-forge-plan
apex-forge-qa		apex-forge-qa
apex-forge-retro		apex-forge-retro
apex-forge-review		apex-forge-review
apex-forge-round		apex-forge-round
apex-forge-security-audit		apex-forge-security-audit
apex-forge-setup-browser-cookies		apex-forge-setup-browser-cookies
apex-forge-setup-deploy		apex-forge-setup-deploy
apex-forge-ship		apex-forge-ship
apex-forge-skill-author		apex-forge-skill-author
apex-forge-unfreeze		apex-forge-unfreeze
apex-forge-verify		apex-forge-verify
apex-forge-wave-challenger		apex-forge-wave-challenger
apex-forge-wave-planner		apex-forge-wave-planner
apex-forge-wave-worker		apex-forge-wave-worker
apex-forge-wave		apex-forge-wave
apex-forge-worktree		apex-forge-worktree
design-to-code		design-to-code
docs		docs
extension		extension
frontend		frontend
hooks		hooks
orchestration		orchestration
platforms		platforms
protocol		protocol
src		src
templates		templates
workflow		workflow
.claude-plugin		.claude-plugin
.gitignore		.gitignore
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
README.zh-CN.md		README.zh-CN.md
SKILL.md		SKILL.md
VERSION		VERSION
labloop.md		labloop.md
mcp-config.json		mcp-config.json
package.json		package.json
setup		setup
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

Apex Forge

Why

Quick Start

Usage

Build a Feature

Fix a Bug

Code Review

QA Testing

Security Audit

More Scenarios

Task Management

Memory System

Browser

Dashboard

All 43 Skills

Protocol (3)

Stages (7)

Quality and Review (6)

Plan Review (4)

Creative (3)

Operations (5)

Safety (4)

Browser (3)

Knowledge (3)

Orchestration (3)

External (2)

CLI Reference

Core Concepts

Complexity Router

Escalation Ladder

Evidence Grading

Verification Gate

TDD Iron Rule

Knowledge Compounding

Installation

Claude Code (global -- recommended)

Claude Code (project-local)

Cursor

Codex / Factory

Gemini / Windsurf

Auto-detect

Configuration

Architecture

Provenance

Troubleshooting

Stats

License

About

Topics

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages