Prompts to Try¶

These prompts work with GitHub Copilot, Cursor, Claude, or any AI coding assistant. Copy one, paste it into your editor's chat, and see what drift finds in your codebase — no config, no setup beyond pip install drift-analyzer.

Not sure what drift even looks for? Check out Problems Every Vibe-Coder Recognizes first — real code examples you'll probably recognize from your own projects.

First Look¶

Start here. One prompt, instant results.

Install drift-analyzer and run `drift analyze --repo .` on this project.
Show me the top 5 findings and explain what each one means for this codebase.

Run `drift analyze --repo .` and give me a plain-language summary:
What's the overall drift score? Which areas of the code have the most structural problems?

Run `drift brief --repo .` and explain the output.
What should I focus on first?

Find Specific Problems¶

Target a specific signal. Each prompt focuses on one type of structural issue.

Duplicate code from AI generation¶

Run `drift analyze --repo . --format json` and look at MDS findings (Mutant Duplicates).
Which functions are near-copies of each other? Show me the pairs and suggest which ones to consolidate.

Architecture boundary violations¶

Run `drift analyze --repo .` and explain all AVS findings (Architecture Violations).
Which imports cross layer boundaries? Draw a simple diagram of the intended layers vs. what drift found.

Pattern fragmentation¶

Run `drift analyze --repo .` and focus on PFS findings (Pattern Fragmentation).
Where is the same concern handled in multiple different ways? Which pattern should be the canonical one?

Complex undocumented code¶

Run `drift analyze --repo .` and check for EDS findings (Explainability Deficit).
Which functions are complex but lack documentation? Prioritize by complexity score.

Fix Workflows¶

Go from findings to fixes. These prompts drive a complete repair loop.

Run `drift fix-plan --repo . --max-tasks 3` and implement the first task from the plan.
Explain what you changed and why. Then re-run `drift analyze --repo .` to confirm the score improved.

Run `drift scan --max-findings 5` and pick the highest-impact finding.
Refactor the code to fix it. Show me a before/after diff and explain the structural improvement.

Run `drift diff --staged-only` on my staged changes.
Are there any new structural problems? If yes, suggest fixes before I commit.

Create a fix plan with `drift fix-plan --repo .` and work through the tasks one by one.
Commit each fix separately with a clear commit message explaining the structural improvement.

CI and Workflow Setup¶

Set up drift in your development workflow — pre-commit, CI, or editor integration.

Pre-commit hook¶

Set up drift as a pre-commit hook in this repo using the report-only mode.
Add the config to .pre-commit-config.yaml and test it with a dry run.

GitHub Actions¶

Create a GitHub Actions workflow file (.github/workflows/drift.yml) that:
1. Runs drift on every pull request
2. Uploads findings as SARIF to GitHub Code Scanning
3. Uses report-only mode (fail-on: none) for now

Use the official drift GitHub Action (mick-gsk/drift@v1).

MCP integration¶

Run `drift init --mcp` to set up MCP integration for this editor.
Then use the drift tools directly — no more copy-pasting prompts needed.

GitLab CI¶

Create a .gitlab-ci.yml job that runs drift in report-only mode on merge requests.
Archive the JSON report as a CI artifact.

Trend and Baseline¶

Track architecture health over time.

Run `drift analyze --repo . --format json -o drift-baseline.json` to create a baseline snapshot.
What's the current drift score? Which signals contribute most?

Run `drift trend --last 30` and explain the trend.
Is the architecture getting better or worse? Which signals are driving the change?

Compare the current drift score against the last baseline.
Run `drift analyze --repo .` and tell me which findings are new since the last snapshot.

Vibe-Coding Survival Kit¶

If your team uses AI assistants heavily, start here for the full workflow.

Copy the vibe-coding config from https://github.com/mick-gsk/drift/tree/main/examples/vibe-coding
into this project's root as drift.yaml. Then:
1. Run `drift validate` to check the config
2. Run `drift analyze --repo .` to get a Day 0 baseline
3. Run `drift fix-plan --repo . --max-tasks 5` and fix the top items

Run `drift analyze --repo .` with a focus on AI-specific patterns:
- MDS: Are there duplicate helpers that an AI assistant generated in multiple places?
- PFS: Is the same concern solved differently across files?
- BAT: Are there `# type: ignore` or broad `except Exception` accumulations?
Show me the findings sorted by impact.

Copy-Ready Workflow Files¶

Complete files you can drop into your project. Copy the content, save at the indicated path, done.

`.github/workflows/drift.yml`¶

.github/workflows/drift.yml

name: Drift
on:
  pull_request:
  push:
    branches: [main]

permissions:
  contents: read
  security-events: write   # only needed if upload-sarif: true

jobs:
  drift:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 50

      - uses: mick-gsk/drift@v1
        with:
          fail-on: none          # report-only — change to 'high' when ready
          upload-sarif: true     # findings appear as PR annotations

`.pre-commit-config.yaml` (add drift hook)¶

.pre-commit-config.yaml

repos:
  - repo: https://github.com/mick-gsk/drift
    rev: v2.6.0
    hooks:
      - id: drift-report        # report-only — switch to drift-check later

`.vscode/mcp.json` (MCP for Copilot / Cursor / Windsurf)¶

.vscode/mcp.json

{
  "servers": {
    "drift": {
      "type": "stdio",
      "command": "drift",
      "args": ["mcp", "--serve"]
    }
  }
}

Shortcut: drift init --mcp generates this file for you.

Pro tip: Use MCP instead of prompts

If your editor supports MCP (VS Code Copilot, Cursor, Claude Desktop, Windsurf), run drift init --mcp once. After that, your AI assistant can call drift tools directly — no prompt copying needed. The prompts above are most useful for quick one-off checks or editors without MCP support.

Vibe-coding optimized config

The examples/vibe-coding/ directory contains a pre-tuned drift.yaml with signal weights and thresholds optimized for AI-heavy codebases, plus a GitHub Actions workflow, pre-push hook, and MCP config — ready to copy into your project.