Skip to content
Back to skills

Framework Detector

ASecurity

Maps a detected stack (from `detect-stack` / atlas) to the framework-specific SKILL.md that should be activated for this session. Invoke once at session start, or whenever a new spec mentions a stack that isn't already loaded. Reads `.claude/memory/atlas/STACK.md` and emits a list of skills to consult.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 9, 2026
ai-agentsrustgobashreactnextjsnodeexpressfastapidjangoflask

Works with

  • claude code
  • api
  • mcp

Security analysis

A100/100

Scanned October 9, 2026

npx -y skills add qyndex/claude-init --skill framework-detector --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Framework Detector?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Framework Detector
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/qyndex-framework-detector/badge)](https://www.skillsdirectory.com/skills/qyndex-framework-detector)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: framework-detector
description: Maps a detected stack (from `detect-stack` / atlas) to the framework-specific SKILL.md that should be activated for this session. Invoke once at session start, or whenever a new spec mentions a stack that isn't already loaded. Reads `.claude/memory/atlas/STACK.md` and emits a list of skills to consult.
when_to_use: When session starts and atlas/manifest.json has detected frameworks; maps detected stack to recommended skills.
disable-model-invocation: true
---

# Framework detector

A routing skill. It does no work itself — it reads the atlas and tells you which framework-specific skill(s) to load.

## Inputs
- `.claude/memory/atlas/STACK.md` — produced by `detect-stack` (`bash .claude/scripts/atlas-refresh.sh`)
- Optional override: spec frontmatter `stack:` field

## Routing table

| Detected `frameworks.web` | Detected `frameworks.api` | Skill to activate |
|---|---|---|
| Vite-React | (none) | `.claude/skills/react-vite/SKILL.md` |
| Vite-React | Express \| Fastify \| Hono \| NestJS | `react-vite` + `react-node` |
| Next.js (App or Pages) | Express \| Fastify \| Hono \| (same Next.js) | `react-node` (+ proposed `nextjs` when seeded) |
| Remix | any Node | `react-node` (+ proposed `remix` when seeded) |
| (any React) | tRPC | `react-node` |
| (none) | Flask | `flask-realtime` if SocketIO present, else proposed `flask-rest` |
| (none) | FastAPI | proposed `fastapi` |
| (none) | Django | proposed `django` |
| (none) | Rails | proposed `rails` |
| (none) | Gin / Fiber / Echo | proposed `go-web` |
| (none) | Axum / Actix / Rocket | proposed `rust-web` |

## How to use

1. **Read the atlas:**
   ```bash
   cat .claude/memory/atlas/STACK.md
   ```
2. **Look up the row** in the routing table.
3. **Reference the skill file by path** in your reasoning — Claude Code's progressive disclosure loads it on demand. Do NOT inline the skill body.
4. **If the row says "proposed":** that skill isn't seeded yet. File a `mcp__ccd_session__spawn_task` to draft it, then proceed with general principles from the relevant `.claude/rules/{frontend,backend}.md`.

## Update to `detect-stack`

When `detect-stack` produces its manifest, it now also writes the *recommended skill list* to `.claude/memory/atlas/SKILL_HINTS.md`:

```markdown
# Recommended framework skills (auto-generated)
- React + Vite detected → load `.claude/skills/react-vite/SKILL.md`
- Flask + SocketIO detected → load `.claude/skills/flask-realtime/SKILL.md`
```

This file is short (10–20 lines), cheap to keep in context, and gives the architect/implementer agents an unambiguous "load this" signal on every session.

## When NOT to invoke
- Single-file scripts, throwaway repos, no `package.json` / `pyproject.toml` — atlas is empty, no routing to do.
- The user explicitly says "ignore the stack, just edit this file".

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.