Argus

the watchman whose eyes never all close

Quick start · What it looks like · Hook scripts · Docs · getargus.org

The hook control center for AI coding agents. Hooks are how you govern what
Claude Code and Codex can do — block dangerous commands, protect secrets, enforce
branch policy, get notified. But managing them means hand-editing JSON, testing
them means waiting for a live agent to misbehave, and good scripts are scattered
across a thousand gists. Argus fixes all three, locally:

Hook management — a config editor with one-click presets for Claude Code and
Codex. No JSON surgery; argus-managed entries are tagged and reversible.
Hook simulator — run any hook command or script against a realistic synthetic
payload for any event type, and inspect stdout/stderr/exit code/duration before
an agent ever fires it. The missing debugger for the hook ecosystem.
Public script collection — my-custom-hook-scripts/
ships battle-tested, zero-dependency guardrails free for everyone: dangerous-command
blocker, secrets protection, branch guard, auto-format with lint feedback,
prompt-injection scanner, webhook notifications, and more. Every script works
with Claude Code and Codex — and the in-app Scripts library lets you browse,
search, and one-click install any of them into ~/.argus/hooks/.
Community sharing — browse and install the whole community registry, and (with
an optional GitHub login) publish your own scripts via a pull request or back
up your collection to a private gist. Nothing is uploaded except the scripts you
choose to share.

Backing it up: a live observability layer — every hook payload is normalized to a
canonical event model, persisted to SQLite, and streamed to a real-time dashboard
(event feed, session explorer, usage and cost stats). You see your hooks — and your
agents — actually working. No cloud, no telemetry, your data stays local.

Quick start

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/ruydt/argus/main/install.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/ruydt/argus/main/install.ps1 | iex

Requirements: Node.js 18+ (plus curl/tar on macOS/Linux) — no Go or pnpm needed.

The installer downloads a pre-built binary for your OS and arch, wires the Claude Code
SessionStart hook, and places argus in ~/.argus/bin.

Open http://127.0.0.1:10804 after your next Claude Code or Codex session starts.

Then follow docs/quickstart.md to verify your first event.

What it looks like

The dashboard is one local panel over your agents: a live event feed, per-project
sessions, token-usage stats, the hooks config editor, and the built-in hook
simulator. Take the full visual tour at getargus.org.

Surface	What it does
Hooks config + simulator	One-click presets; fire a synthetic payload at any hook and read stdout/stderr/exit code before a live agent runs it.
Event feed	Every normalized tool call streamed over SSE, with server-side search across session id and project.
Sessions	Per-project session table — duration, tokens, agent, timing — with a file-change drawer of per-edit diffs.
Dashboard	Token-usage roll-ups (input/output/cache) charted over time and by model — computed locally.
Projects	Cards per working directory, server-searchable with infinite scroll and cascade delete.
Diagnostics	Health, storage, hook-preset detection, `~/.argus` inventory, and log tails.
Scripts library	Browse community scripts, inspect source in a modal, install into `~/.argus/hooks/`, and manage local/gist copies from My Collection.
Script collection	Free zero-dependency guardrails in `my-custom-hook-scripts/`.

Uninstall

curl -fsSL https://raw.githubusercontent.com/ruydt/argus/main/uninstall.sh | bash

Stops the server, removes binaries and scripts, unwires hooks from ~/.claude/settings.json, and optionally deletes your data.

Documentation

docs/hooks.md - hook management, presets, and the hook simulator
my-custom-hook-scripts/ - the public hook script collection
docs/quickstart.md - first-event walkthrough (under 10 minutes)
docs/install.md - full install reference, support matrix, data lifecycle
docs/privacy.md - capture categories, ignore controls, export implications
docs/security.md - local threat model and remote-sharing posture
docs/releases.md - release runbook and conventional commit format

Contributing

See CONTRIBUTING.md.

License

LICENSE