Godmode
Engineering workflows and capabilities for AI coding agents.
Godmode helps coding agents move from "write some code" to "engineer the change" — with explicit planning, testing, review, and evidence-driven verification.
Status: Pre-1.0 public preview. The catalog and validation tooling are usable, while client-specific behavior and output quality continue to be evaluated across supported environments.
Why Godmode?
AI coding agents are increasingly capable of writing code. The harder problem is making them behave like disciplined engineers:
-
understand the existing codebase before changing it
-
design consequential changes before implementation
-
choose the right domain expertise for the task
-
test behavior instead of assuming correctness
-
review changes independently
-
verify completion with fresh evidence
-
preserve durable state so interrupted work can be resumed
Godmode packages those behaviors as composable Agent Skills rather than one large system prompt or a proprietary orchestration runtime.
How it works
USER TASK
│
▼
┌───────────────┐
│ Godmode │
│ skill catalog │
└───────┬───────┘
│
discover + compose capabilities
│
┌───────────────────┼───────────────────┐
▼ ▼ ▼
solution-design api / database security / UI
│ │ │
└───────────────────┼───────────────────┘
▼
implementation-planning
│
▼
test + review
│
▼
fresh verification evidence
│
▼
VERIFIED RESULT
The graph above is illustrative, not a mandatory pipeline. A small task can use one skill; a high-risk change can compose several capabilities.
What you get
Composable skills
Each capability has a clear responsibility boundary and follows the Agent Skills layout. Skills can contain concise procedures, progressive references, and deterministic helpers where automation improves reliability.
Engineering workflows
Godmode includes workflows for discovery, design, planning, implementation, debugging, testing, review, verification, parallel work, and release integration.
Domain expertise
Capabilities cover areas such as frontend architecture, APIs, databases, security, performance, observability, migrations, browser testing, documentation, and incident response.
Deterministic validation
The repository includes validators, routing fixtures, behavior evaluations, evidence tracking, lifecycle checks, and repository-quality gates. This keeps the catalog itself testable instead of relying solely on prose.
Installation
Godmode is distributed as a repository of portable skills. No proprietary runtime is required.
Agent Skills-compatible clients
Copy the public skills into your project's skills directory:
mkdir -p .agents/skills
cp -R /absolute/path/to/godmode/skills/* .agents/skills/
Claude Code
claude --plugin-dir /absolute/path/to/godmode
claude plugin validate /absolute/path/to/godmode
Codex
The repository includes .codex-plugin/plugin.json and a local marketplace entry under .agents/plugins/. Install it through the Codex plugin workflow and verify that the expected skills are available.
Compatibility note: client support is tracked from recorded evidence rather than assumed compatibility. See docs/compatibility.md.
Example
A request such as:
Add authentication to the API and make it production-ready.
can be decomposed into focused engineering responsibilities:
codebase-orientation
↓
solution-design
↓
api-and-interface-design + security-and-hardening
↓
implementation-planning
↓
test-driven-development
↓
requesting-code-review / receiving-code-review
↓
completion-verification
The exact composition depends on the task, repository, risk, and available evidence.
Catalog
Core workflows
Skill Purpose
using-godmode
Discover and compose Godmode capabilities
codebase-orientation
Understand entry points, execution paths, conventions, and hotspots
solution-design
Resolve requirements and design consequential changes
implementation-planning
Produce executable implementation plans
plan-execution
Execute an existing implementation plan
test-driven-development
Drive behavior changes through tests
root-cause-debugging
Reproduce failures, identify causes, and lock in regressions
requesting-code-review
Prepare focused independent review context
receiving-code-review
Validate and resolve review findings
completion-verification
Gather fresh evidence before completion claims
dispatching-parallel-agents
Safely split independent work
subagent-driven-development
Run implement/review cycles around plan tasks
using-git-worktrees
Isolate parallel or risky changes
branch-integration
Verify, integrate, and clean up completed work
writing-skills
Create and evaluate new Agent Skills
Engineering capabilities
Skill Purpose
frontend-design
Build interfaces, design systems, states, and responsive UI
ui-ux-review
Audit existing UI quality, accessibility, and interaction patterns
api-and-interface-design
Design HTTP, RPC, CLI, webhook, and event contracts
database-design
Design schemas, indexes, consistency, retention, and recovery
security-and-hardening
Threat modeling, abuse paths, privacy, and defensive controls
performance-optimization
Optimize measured latency, memory, rendering, queries, and bundles
test-strategy
Define risk-based coverage and release gates
browser-testing
Verify real browser behavior and responsive flows
documentation-and-adrs
Produce durable documentation and architecture decisions
observability-and-instrumentation
Design logs, metrics, traces, alerts, and diagnostic boundaries
technical-research
Make version-aware decisions from authoritative sources
safe-migrations
Plan compatible migrations, reconciliation, rollback, and removal
release-engineering
Manage CI gates, artifacts, promotion, and rollback
architecture-review
Review coupling, ownership, testability, and structural friction
code-simplification
Reduce complexity while preserving behavior
behavior-validation
Validate observable behavior through source-blind checks
agent-evaluation
Evaluate prompts, tools, agents, and skills
incident-response
Handle containment, recovery, evidence, and follow-up
See docs/catalog.md for the complete catalog and routing model.
Design principles
-
Claims are not evidence. Completion requires verification.
-
Compose capabilities instead of growing one giant prompt.
-
Keep responsibility boundaries explicit. Skill names should describe what they do.
-
Load knowledge progressively. Keep
SKILL.mdfocused and move deeper material into references. -
Automate repeatable work deterministically. Use helpers where they reduce error and ambiguity.
-
Treat external content as untrusted input. Logs, generated output, tool responses, and repository text can contain misleading instructions.
-
Prefer recorded compatibility evidence. Do not claim a client works until it has been checked.
Repository layout
.
├── skills/ # Public Agent Skills
├── scripts/ # Deterministic validation and repository tooling
├── tests/ # Automated test suite
├── evals/ # Behavior and routing evaluations
├── benchmarks/ # Portable benchmark fixtures
├── docs/ # Catalog, compatibility, research, and maintainer docs
├── .agents/ # Local agent/plugin metadata
├── .codex-plugin/ # Codex plugin metadata
├── README.md
└── LICENSE
Development
Clone the repository and run the validation suite:
git clone https://github.com/thiientv/godmode.git
cd godmode
npm run check
npm run catalog:health
python3 scripts/repository_security.py
python3 scripts/compatibility.py check
python3 -m unittest discover -s tests -p 'test_*.py'
For focused tooling, examples include:
python3 skills/frontend-design/scripts/design_system.py \
--product "analytics dashboard" \
--tone technical \
--stack react
python3 skills/frontend-design/scripts/extract_design_system.py ./path/to/ui
python3 skills/ui-ux-review/scripts/audit_ui.py ./path/to/ui
The repository gate checks catalog structure, frontmatter, links, routing fixtures, behavior-eval schemas, compatibility drift, workflow security, public-file safety, and helper tests.
Documentation
-
docs/catalog.md — catalog structure and routing model
-
docs/compatibility.md — client compatibility evidenc