# Stop Your AI Agent From Building Code on Abandoned 2022 Libraries (Open-Source MCP Server) > Source: > Published: 2026-10-28 14:34:05+00:00 **Intelligent technology stack and open-source curator for AI coding agents.** *Never let your AI agent build production code on abandoned 2022 libraries.* When you ask Cursor, Claude Code, Antigravity, or Copilot: *"Build an AI meeting assistant that records audio, separates speakers, and generates summaries"* The model immediately starts writing code using libraries from its **static pre-training data**. Meanwhile: `GPL-3.0` or `CC-BY-NC-4.0` libraries can create major intellectual property and commercial liabilities. **`skills-but-for-dev` fixes this.** It acts as an autonomous pre-flight curator for your agent, querying live GitHub velocity, PyPI monthly adoption, and Hugging Face VRAM requirements **before** writing code. | Task: *"Build a local meeting notes agent"* | ❌ Vanilla AI Agent (Without `skills-but-for-dev` ) | ✅ AI Agent + `skills-but-for-dev` | |---|---|---| | **Transcription (ASR)** | Suggests `SpeechRecognition` (deprecated Python 2 era wrapper) | **`Systran/faster-whisper`** (9.6M/mo PyPI downloads, 4x faster CTranslate2) | | **Speaker Diarization** | Tries to hand-code clustering or hallucinates non-existent APIs | **`modelscope/FunASR`** /**` pyannote.audio`** (Verified active acceleration: $v_4 / v_{12} = 1.2$) | | **Licensing** | Generates code importing `GPL` / non-commercial packages | **100% Permissive Verification** (Flags MIT / Apache-2.0, filters restrictive licenses) | | **Local Hardware Check** | Recommends large Whisper model, crashing local GPU | **Automatic VRAM Baseline** (Estimates 0.54 GB FP16 ceiling for`faster-whisper-small` ) | `stars:>500`) alongside fast-rising breakout repositories (` stars:75..3500`, pushed $< 30$ days) without cannibalization.` pypistats.org` to protect against star-inflation and GitHub bots.`.safetensors`, sharded model weights, Diffusers pipelines, and distinct GGUF quantization levels (`Q4_K_M`, `Q8_0`).` gh auth token`) or falls back to an adaptive, quota-conserving unauthenticated mode. ``` flowchart TD User["Developer Prompt\n'Build a RAG system for PDF invoices'"] --> Agent["AI Coding Agent\n(Cursor / Antigravity / Claude Code)"] Skill["SKILL.md"] -.->|"Decomposition Instructions"| Agent Agent -->|"1. search_github_repos('pdf parser')\n2. search_hf_models('embeddings')"| Server["skills-but-for-dev MCP Server"] subgraph Server_Internals["skills-but-for-dev Engine"] Server --> Norm["Query Normalizer\n(Noise stripping & Language filter)"] Norm --> GH["GitHub REST Client\n(Dual-Bucket search + Star Velocity)"] Norm --> HF["Hugging Face Client\n(VRAM & Pipeline Estimator)"] Norm --> PyPI["PyPI Stats Client\n(30-Day Download Verification)"] GH --> Scoring["Conservative Scoring Engine\n(80% Base + 10% Health + 10% PyPI)"] HF --> Scoring PyPI --> Scoring Scoring --> Cache[("SQLite Cache (WAL Mode)\n~4.7ms warm latency")] end Server -->|"Curated Architecture Recipe"| Agent Agent -->|"Step-by-step Tech Stack with Verified Licences"| User ``` `uvx` (Zero Installation) No cloning or virtual environment setup required. Works with any MCP-compatible client: ``` # Verify it runs and check diagnostics uvx skills-but-for-dev ping ``` `.cursor/mcp.json`) ``` { "mcpServers": { "skills-but-for-dev": { "command": "uvx", "args": ["skills-but-for-dev"], "env": { "GITHUB_TOKEN": "optional_your_pat_here" } } } } ``` `mcp_config.json`) ``` { "mcpServers": { "skills-but-for-dev": { "command": "uvx", "args": ["skills-but-for-dev"] } } } ``` `claude_desktop_config.json`) ``` { "mcpServers": { "skills-but-for-dev": { "command": "uvx", "args": ["skills-but-for-dev"] } } } ``` **Tip:** If you already use the GitHub CLI (`gh`), `skills-but-for-dev` automatically detects your authentication token via `gh auth token`. You don't even need to configure `GITHUB_TOKEN` manually! When your agent uses `skills-but-for-dev`, it generates structured architecture recipes before writing code: | Layer | Recommended Component | Star Velocity | Real-World PyPI Adoption | License | Hardware / Memory | |---|---|---|---|---|---| | **ASR (Speech-to-Text)** | [`SYSTRAN/faster-whisper`](https://github.com/SYSTRAN/faster-whisper) | $v_4 = 18.2$/day (accel: 1.15) | **9,614,222** downloads/mo | MIT | ~0.54 GB VRAM (small FP16) | | **Speaker Diarization** | [`modelscope/FunASR`](https://github.com/modelscope/FunASR) | $v_4 = 8.4$/day (accel: 1.20) | **272,534** downloads/mo | MIT | ~1.2 GB VRAM | | **Embeddings** | [`sentence-transformers`](https://github.com/UKPLab/sentence-transformers) | $v_4 = 14.1$/day (accel: 1.05) | **22,332,616** downloads/mo | Apache-2.0 | ~0.62 GB VRAM | Benchmarked across realistic multi-layer agent scenarios using strict Ground Truth matching: | Metric | Result | Benchmark Details | |---|---|---| | **Warm Cache Latency** | **4.73 ms** | Sub-5 millisecond response time via SQLite WAL mode | | **Cold API Latency** | **1,728 ms** | Optimized single-flight queries with adaptive quotas | | **Cache Speedup** | **365.7x** | Massive latency reduction for recurring architecture queries | | **Commercial License Safety** | **95.2%** | Automatic filtering of GPL / CC-BY-NC restrictive licenses | | **PyPI Download Signal Rate** | **52.4%** | Over half of returned candidates enriched with production downloads | | **Unit Test Coverage** | **40 / 40 Passing** | Comprehensive suites for normalizer, PyPI client, cache, and query relaxation | You can run the evaluation suite locally at any time: ``` python evals/run_evals.py ``` Inspect active tokens, cache paths, and server health directly from your terminal: ``` # Print active version skills-but-for-dev --version # Ping server and verify token discovery source (ENV vs GitHub CLI) skills-but-for-dev ping ``` For mathematical formulas (star velocity moving averages, exponential recency decay, dormant spike detection ratios), see [ARCHITECTURE.md](//ARCHITECTURE.md). Distributed under the [Apache License 2.0](https://dev.toLICENSE).