HELMOR Agent banner

HELMOR Agent

The local agent watcher that keeps AI-assisted product development aligned, safer, and easier to hand off.

HELMOR gives Codex, Claude Code, Cursor, Windsurf, and other AI coding agents a project-aware control layer for context, 140+ skill routing, detector checks, and release discipline.

89% token savings
target

95% less hallucinations
target

Zero drift
truth-file alignment

140+ skills
agent routing library

Guardrails
best-in-class UI/UX and safety review

Install

Platform	Command
npm	`npm i -g @helmoragent/helmor`
one-off npm	`npx @helmoragent/helmor@latest install`
pnpm	`pnpm dlx @helmoragent/helmor install`
yarn	`yarn dlx @helmoragent/helmor install`
bun	`bunx @helmoragent/helmor install`
macOS Homebrew	`brew install helmorx/tap/helmoragent`
Windows PowerShell	`irm https://raw.githubusercontent.com/helmorx/helmoragent/main/install/install.ps1 \| iex`
Linux shell	`curl -fsSL https://raw.githubusercontent.com/helmorx/helmoragent/main/install/install.sh \| sh`

Quickstart

helmor install
helmor status
helmor dashboard
helmor doctor

helmor install now writes the project profile/adapters and merges official Codex/Claude global hook entries. Use helmor install --project-only when you want reviewable project files without changing global hooks.

Existing projects start in observe mode, so HELMOR warns, routes, and summarizes before it blocks. Move to guard or strict when the repo is ready for stronger enforcement.

Why HELMOR

AI agents are fast, but they waste tokens rediscovering the same repo, forget decisions, invent missing APIs, run the wrong commands, and drift away from product truth. HELMOR sits inside each project as a local agent layer for the whole development lifecycle: fewer repeated context reads, fewer hallucination-prone turns, stronger UI/UX review discipline, and safety guardrails before handoff or release.

Context

Start sessions with compact repo state, project profile, and handoff context.

Routing

Send prompts toward the right built-in skill before an agent spends tokens.

Safety

Watch commands, files, package runners, secrets, and release-sensitive paths with configurable guardrails.

Closeout

Track checks, findings, touched areas, handoff notes, and next-agent continuity.

How It Works

HELMOR Agent workflow diagram

flowchart LR
  A[AI coding agent] --> B[HELMOR Agent]
  B --> C[Project profile]
  B --> D[Skill router]
  B --> E[Detector packs]
  B --> F[Checks and handoff]
  C --> G[Aligned implementation]
  D --> G
  E --> G
  F --> G

Session Dashboard

HELMOR terminal dashboard preview

Lifecycle Graph

flowchart LR
  Plan[Plan] --> Build[Build]
  Build --> Watch[Watch]
  Watch --> Verify[Verify]
  Verify --> Ship[Ship]
  Ship --> Remember[Remember]
  Remember --> Plan

  Plan -.-> P1[Product intent, scope, architecture]
  Build -.-> B1[Frontend, backend, data, infrastructure]
  Watch -.-> W1[Context, tools, package runners, unsafe actions]
  Verify -.-> V1[Tests, security review, detector findings]
  Ship -.-> S1[Release blockers and launch readiness]
  Remember -.-> R1[Handoffs and next-agent continuity]

Signal Graph

HELMOR Agent signal graph

Signal	HELMOR watches	Result
repo truth	docs, project profile, adapters, checks	less drift from the intended product
tool use	shell commands, git actions, package runners	earlier warnings or blocks for risky actions
project state	active task, touched areas, findings	cleaner handoff and closeout
verification	tests, doctor output, detector packs	repeatable pre-release evidence

What It Adds To A Project

.helmor/
  project.json          repo profile, checks, policies, tools, adapters
  context-card.md       compact context for new sessions
  handoff.md            closeout summary for the next agent
  state.json            local runtime state, ignored by git

HELMOR is local-first. It does not require an account, upload source, or send telemetry in v1.

Agent Support

Agent	V1 support	Integration style
Codex	yes	hook-compatible command entrypoints
Claude Code	yes	hook-compatible command entrypoints
Cursor	yes	generated project rules
Windsurf	yes	generated project rules
Other agents	compatible	`helmor hook --event <EventName>`

Modes

Mode	Use it when	Behavior
`observe`	adopting HELMOR in an existing repo	warn, route, summarize
`guard`	active development with agents	block secrets, destructive git, wrong runner, unsafe deploys
`strict`	release, launch, security-sensitive work	enforce checks, handoffs, closeout, security review

HELMOR Skills

HELMOR ships with 140+ built-in skills that route AI coding agents toward the right behavior before they spend tokens or touch code.

Lifecycle	Skills
Plan	Product Planning, Architecture, API Contracts
Build	Frontend, Backend, Data, Infrastructure, UI Design
Verify	Testing, Security, Launch Readiness
Remember	Project Memory, Token Reduction, Docs & Handoff

Docs

Topic	Link
Quickstart	docs/QUICKSTART.md
Skills	docs/SKILLS.md
Commands	docs/COMMANDS.md
Agent adapters	docs/ADAPTERS.md
Detector packs	docs/DETECTORS.md
Project profile	docs/PROJECT_PROFILE.md
Security policy	SECURITY.md
Contributing	CONTRIBUTING.md
Changelog	CHANGELOG.md

Built For

developers building new products with AI agents
vibe coders who need less hallucination and more structure
teams that want repeatable AI-assisted development workflows
projects that need context, testing, handoff, and launch discipline

License

Apache-2.0. See LICENSE.