# intent-engineering

Scores agent intent specs against a 25-item rubric, builds new ones from a nine-section template, and says how much an existing prompt needs reworking.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI
- Author: seanwinslow28
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/seanwinslow28/sw-mcp-intent-engineering
- Updated: 2026-08-29
- HTML: https://affpapa.org/en/nevibecode/mcp/intent-engineering

## What it does

Three tools come with the server. `audit_intent_spec` takes inline text or a file path and hands back a score out of 25, findings per section, anti-patterns it spotted and three fixes to make first. `generate_intent_spec_scaffold` writes a YAML skeleton for the chosen kind: blank, level-1-mvr or full-9-section. `assess_retrofit_level` reads an existing prompt or `SKILL.md` and picks L1, L2 or L3, with reasoning about complexity and how much autonomy the agent gets.

The rubric, the five fatal anti-patterns, the four autonomy levels and the nine-section layout all trace to the canonical `intent-engineering` skill. This is an adapter around it, not a fork. Every tool is read-only: nothing gets written to disk. Disk reads go through one guard — an extension allowlist, symlink resolution, confinement to a root, and a size cap. An earlier build would open whatever path it was given, so that guard closes a real disclosure bug.

You need Node 20+ and a client that speaks MCP. Transport is stdio only, which ties the server to a process-bound client and rules out cloud agents for now; the author plans SSE in version 1. Heading parsing is opinionated and its alias table is documented. One honest limit: the mapper recognizes an equivalent intent already in the text, it does not invent one.

## Requirements
- Needs: Node.js (commands run via npx)

## Install

### Claude Code (native platform)

**Platform CLI**
```bash
claude mcp add intent-engineering -- npx -y @swins/intent-engineering-mcp
```

### Codex

**Platform CLI**
```bash
codex mcp add intent-engineering -- npx -y @swins/intent-engineering-mcp
```

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "intent-engineering": {
            "command": "npx",
            "args": [
                "-y",
                "@swins/intent-engineering-mcp"
            ]
        }
    }
}
```

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "intent-engineering": {
            "type": "stdio",
            "command": "npx",
            "args": [
                "-y",
                "@swins/intent-engineering-mcp"
            ]
        }
    }
}
```

### Gemini CLI

**Platform CLI**
```bash
gemini mcp add intent-engineering npx -y @swins/intent-engineering-mcp
```

## When to use it
- catch a vague objective before an agent acts on it
- get section-by-section findings instead of a single grade
- scaffold a new intent spec at the right level of detail
- decide whether an old prompt or `SKILL.md` is worth rewriting
- check a spec for fatal anti-patterns

## Example prompts
- Run `audit_intent_spec` on `specs/checkout-agent.md`
- Generate a scaffold with `generate_intent_spec_scaffold`, kind level-1-mvr
- Use `assess_retrofit_level` on this `SKILL.md` and tell me if it needs rework

## Similar
- [intent-modeling](https://affpapa.org/en/nevibecode/skills/intent-modeling.md) — skill · 527 installs
- [context-engineering-advisor](https://affpapa.org/en/nevibecode/skills/context-engineering-advisor.md) — skill · 2700 installs
- [prompt-engineering](https://affpapa.org/en/nevibecode/skills/neolabhq-prompt-engineering.md) — skill · 1400 installs
- [prompt-engineering-instructor](https://affpapa.org/en/nevibecode/skills/prompt-engineering-instructor.md) — skill · 198 installs
- [prompt-engineering-patterns](https://affpapa.org/en/nevibecode/skills/prompt-engineering-patterns.md) — skill · 23418 installs
- [ai-prompt-engineering-safety-review](https://affpapa.org/en/nevibecode/skills/ai-prompt-engineering-safety-review.md) — skill · 10327 installs

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md