Skip to main content
Glama

PromptShift ⚔ Universal AI Coding Agent Transpiler & Drift Guard

PromptShift Certified GitHub release License: MIT AST Parity PromptShift MCP server Zero-Data-Retention MCP Compatible

Deterministic AST transpilation and secret sanitation between Cursor (.mdc), Claude Code (CLAUDE.md), Windsurf (.windsurfrules), GitHub Copilot, Cline, and Roo Code.

🌐 Web Application: https://promptshift.dev
šŸ“¦ CLI Package: npx promptshift
šŸ¤– MCP Server: npx promptshift --mcp
šŸ”„ GitHub Action: topaisaas-dev/promptshift@v1


The Problem: AI Rule Fragmentation & Multi-IDE Drift

Modern engineering teams often collaborate across multiple AI-assisted developer environments:

  • Cursor uses granular frontmatter-annotated markdown rules (.cursor/rules/*.mdc) and .cursorrules.

  • Claude Code expects a unified root instruction context (CLAUDE.md).

  • Windsurf relies on .windsurfrules.

  • GitHub Copilot looks for .github/copilot-instructions.md.

  • Aider, Cline & Roo Code each define custom agent boundaries.

When an engineer updates CLAUDE.md but forgets to synchronize .cursor/rules/, AI coding agents develop conflicting behaviors, hallucinate mismatched constraints, and accidentally commit sensitive API keys.

PromptShift eliminates fragmentation with certified 98.5% AST semantic parity, OWASP In-RAM secret sanitization, and instant CI/CD drift guards.


Related MCP server: Fetch MCP Server

⚔ Quick Start

1. One-Command Continuous CI/CD Guard

Bootstrap automated PR parity verification and secret scanning in 2 seconds:

npx promptshift --init-ci

2. Zero-Install CLI (NPX)

Transpile rules locally without permanent installation:

# Real-time watcher: automatically syncs rules across Cursor & Claude on every Ctrl+S (0.02s)
npx promptshift --watch

# Audit workspace for rules drift and exposed secrets
npx promptshift --check

# Transpile from current project to Claude Code
npx promptshift --to=claude

# Transpile from current project to GitHub Copilot
npx promptshift --to=copilot

# Transpile from current project to Windsurf
npx promptshift --to=windsurf

# Get official certified badge for your README.md
npx promptshift --badge

2. Model Context Protocol (MCP) Integration

PromptShift includes a native STDIO JSON-RPC 2.0 MCP server compatible with Claude Desktop, Cursor, Glama.ai, and Smithery.ai.

Claude Desktop Configuration (claude_desktop_config.json)

{
  "mcpServers": {
    "promptshift": {
      "command": "npx",
      "args": ["-y", "promptshift", "--mcp"]
    }
  }
}

Docker Container Execution (Glama Firecracker & Local MicroVMs)

# Build and run the lightweight MCP server via Docker
docker build -t promptshift .
docker run -i --rm promptshift

Exposed MCP Tools (TDQS 5-Star Grade A):

  • promptshift_transpile: Deterministically compiles rules between Cursor, Claude Code, Windsurf, Copilot, Cline, and Aider with 98.5% AST parity.

  • promptshift_sanitize_secrets: Automatically redacts exposed OpenAI, Anthropic, AWS, GitHub, and Stripe API keys in-RAM before commits.

  • promptshift_validate_ast: Measures semantic AST fidelity and prompt drift percentage between source and target instruction documents.


šŸ”„ GitHub Actions CI/CD Drift Guard

Add PromptShift to your pull request workflows to automatically enforce rule synchronization across your team:

# .github/workflows/ai-rules-sync.yml
name: AI Agent Rules Synchronization & Drift Guard

on:
  pull_request:
    paths:
      - 'CLAUDE.md'
      - '.cursor/rules/**'
      - '.windsurfrules'

jobs:
  sync-rules:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout repository
        uses: actions/checkout@v4

      - name: Run PromptShift AST Sync
        uses: topaisaas-dev/promptshift@v1.1.0
        with:
          from: 'cursor'
          to: 'claude-code'
          license_key: ${{ secrets.PROMPTSHIFT_LICENSE_KEY }}

šŸ”’ Security & OWASP Secret Sanitizer

All transpilation processes run 100% in-memory with zero disk persistence and zero telemetry. Before converting instructions, PromptShift runs an OWASP heuristic regex scanner to sanitize:

  • OpenAI API Keys (sk-proj-...)

  • Anthropic API Keys (sk-ant-...)

  • GitHub Tokens (ghp_..., github_pat_...)

  • AWS Access Keys (AKIA...)

  • Stripe API Keys (sk_live_..., rk_live_...)

  • Private RSA / OpenSSH Cryptographic Headers

Leaked credentials are substituted in-place with [REDACTED_API_KEY].


šŸ’³ Pricing & Commercial Access (Polar.sh)

  • Free Starter: Up to 2 rule files completely free forever (1 master context + 1 specialized rule).

  • Single Deposit Pass ($1.99 one-time): Lifetime unlock for a single repository migration. šŸ‘‰ Direct Checkout Link

  • Pro Developer ($5.99/mo): Unlimited rule files, priority AST compilation, and CLI unlocks. šŸ‘‰ Direct Checkout Link

  • Team CI/CD & Agency ($19.99/mo): Multi-seat license keys for GitHub Actions CI/CD drift guards. šŸ‘‰ Direct Checkout Link


šŸ“š Citation & Academic / AI References

If you use or reference PromptShift in your research, tools, or articles, please cite:

@software{promptshift2026,
  author = {Garcin, Jean-Yves and TopAI SaaS Team},
  title = {PromptShift: Universal AST Transpiler & Synchronization Bridge for AI Coding Rules},
  url = {https://promptshift.dev},
  year = {2026},
  version = {1.2.0}
}

šŸ“„ License & Compliance

Distributed under the MIT License. Compliant with OWASP Top 10 API Security guidelines, SOC-2 Zero Data Retention principles, and WCAG 2.1 AA accessibility standards.

Available Tools

3 tools
promptshift_sanitize_secretsA

Inspect and redact exposed private credentials (OpenAI sk-, Anthropic sk-ant-, AWS AKIA*, GitHub ghp_*, Google AIza*, Stripe sk_*) from prompt instructions in-RAM. Use this tool before committing, sharing, or deploying AI prompts to prevent credential leaks. Do not use for cryptographic key generation or data encryption. Pure idempotent scanner executing with zero persistent retention.

Inputs: text (required).

promptshift_transpileA

Transpile agent instructions, master prompts, and modular rules across Cursor (.cursorrules / .cursor/rules/*.mdc), Anthropic Claude (CLAUDE.md / Claude Projects XML), OpenAI Assistants, and Google Gemini. Use this tool when migrating or porting AI coding configurations between different developer environments. Do not use for general natural language translation or non-AI code refactoring. Operates 100% in-memory with deterministic AST mapping and active secret sanitization.

Inputs: system_prompt (required), source_platform (required), target_platform (required), rules, +1 more.

promptshift_validate_astA

Evaluate semantic AST fidelity and measure prompt drift percentage between an original agent instruction document and its transpiled counterpart. Use this tool to verify cross-IDE parity and detect omitted constraints after rule conversion. Do not use for code compilation or syntax execution. Returns a normalized fidelity score (0-100%).

Inputs: source_text (required), target_text (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 3 tools

    Disambiguation5/5

    Each tool targets a distinct stage of the workflow: conversion (transpile), credential redaction (sanitize_secrets), and fidelity verification (validate_ast). The descriptions explicitly exclude adjacent use cases (e.g., no general translation, no cryptography), leaving no meaningful overlap.

    Naming Consistency5/5

    All three tools follow the same `promptshift_` prefix plus a clear snake_case verb_noun pattern (transpile, sanitize_secrets, validate_ast). The convention is predictable and readable throughout.

    Tool Count4/5

    Three tools form a tight, well-scoped pipeline for prompt transpilation, sanitization, and validation, with no redundant entries. It is on the lean side, but each tool clearly earns its place and the narrow domain does not demand more.

    Completeness4/5

    The surface covers the core lifecycle: convert prompts between environments, clean secrets, and verify parity afterward. Minor gaps exist around discovering supported source/target formats or retrieving detailed diffs, but agents can work around these from the descriptions.

    Maintenance

    ActivityActive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers