Long-running Agent

Initializer + Coding Agent

Core idea: split one Agent into two roles — one responsible for planning, one for execution. Every handoff leaves a clean state.

Dual-role Solution

Two Agents, clear division of responsibility

Runs on first round only

Initializer Agent

Responsible for bootstrapping: set up environment, define the plan, make the first commit

Create init.sh script to set up the dev environment
Write claude-progress.txt progress file
Expand user's high-level prompt into a detailed feature checklist (JSON format)
Make the first git commit, ensuring the repo is in a clean state

Runs every round

Coding Agent

Responsible for continuous progress: implement features one by one, keep moving forward

Read the progress file to understand the current state
Do one feature at a time
Update the progress file after completion
git commit with a clear message of what was done

Feature Checklist Design

A proven approach is to use JSON format for the feature checklist — Markdown is not suitable for this purpose. The reason: models are less likely to accidentally modify structured JSON, whereas Markdown tends to get rewritten by the model.

// feature checklist in claude-progress.txt
{
  "features": [
    {
      "category": "authentication",
      "description": "Email/password login with session management",
      "steps": [
        "Create login form component",
        "Implement auth API endpoint",
        "Add session cookie handling",
        "Write end-to-end test"
      ],
      "passes": false
    },
    {
      "category": "chat",
      "description": "Real-time streaming chat with Claude API",
      "steps": ["..."],
      "passes": false
    }
  ]
}
  

Use strong wording in the Prompt: explicitly tell the Agent "do not delete or modify existing test content." Otherwise the Agent will lower test standards just to make tests pass.

Incremental Progress: One Feature at a Time

Why "one at a time" is the key

1 After completing each feature, the code is in a mergeable state: no half-baked work, no syntax errors
2 Git commits provide rollback points: if the next round breaks something, you can return to the last clean state
3 The progress file provides context: a new Agent doesn't have to guess where things are — just read the file
4 The context window won't overflow: each round only needs context for one feature, preventing accumulation to the limit

Test Verification

Agents often think they're done but haven't done end-to-end verification. It says "login feature is implemented," but the button doesn't actually click. Solution:

Explicitly require the Agent to do end-to-end testing with browser automation.
You need to actually open a browser, click buttons, and verify results — unit tests alone are not enough. Have the Agent write E2E tests with Puppeteer / Playwright as the criterion for whether a feature truly "passes."

Final Result

200+ feature claude.ai clone successfully built

Using the Initializer + Coding Agent dual-role pattern, the Agent successfully autonomously built a complete web application with 200+ features. Every feature has a corresponding E2E test, and the code always remains in a mergeable state.

Good handoff mechanism = good long-running Agent. Progress files, feature checklists, incremental commits — these are the most basic guarantees for an Agent to keep moving forward. Nothing fancy.