Argus for Claude Code: Cut Repetitive Workflow Token Usage by 80% with

Argus enforces deterministic workflows for Claude Code, cutting token usage ~80% by replacing AI reasoning with structured execution paths. Install in minutes.

What Changed — Argus Brings Deterministic Workflows to Claude Code

Argus is a new open-source agent skill that transforms how Claude Code handles repetitive workflows. Instead of letting the AI reason through every step of a task it's done a hundred times before, Argus enforces a pre-defined execution path. The result? Up to 80% fewer tokens on repetitive tasks, with no loss in accuracy.

Measured across four use cases (see their benchmark), Argus replaces the token-heavy guesswork of AI reasoning with a structured flow. It tracks state changes and only feeds the agent the exact context needed for the current step — no more dumping entire files or conversation histories into the context window.

What It Means For You — Concrete Impact on Daily Claude Code Usage

If you use Claude Code for any of these, Argus directly saves you money and time:

• CI/CD pipelines — Running the same linting, testing, or deployment steps across multiple projects
• Code review workflows — Consistent analysis patterns applied to every PR
• Refactoring tasks — Applying the same structural changes across a codebase
• Documentation generation — Following the same process for every API endpoint

Instead of Claude Code re-reasoning through "how do I lint this project" each time, Argus says: run step 1, then step 2, then step 3. The AI only activates when a step requires genuine reasoning — like choosing a fix strategy. Everything else is deterministic.

Try It Now — Install and Run Your First Workflow

Installation

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

Then initialize in your project folder:

botcircuits init

This creates .botcircuits/settings.json and installs the workflow skills into ~/.claude/skills.

Authoring a Workflow

Describe your process in plain language to Claude Code. The authoring skill writes the workflow JSON for you. For example:

"Create a workflow that lints all changed files, runs unit tests, then reports coverage. Only reason about fixing lint errors — the rest is deterministic."

Argus generates the flow structure. You can also write the JSON directly if you prefer.

Running a Workflow

Once authored, trigger execution:

botcircuits run <workflow-name>

Argus dispatches each step to Claude Code, providing only the context needed for that step. The AI never sees the full workflow — just the current task.

Advanced Configuration

Override runtime settings in .botcircuits/settings.json:

{
  "runtime": "claude-code",
  "runtimes": {
    "claude-code": {
      "command": ["claude", "-p", "{prompt}", "--no-auto-save"],
      "timeout": 300
    }
  }
}

{prompt} is the placeholder Argus substitutes with each step's segment prompt. Use --link during install to symlink skills and auto-pickup updates:

botcircuits init --link

Pro Tip: Symlink for Zero-Maintenance Updates

botcircuits init --link --runtime claude-code

Now when Argus updates, your Claude Code skills update automatically. No re-install needed.

---

Source: github.com

---

Originally published on gentic.news