Sfp

---
name: sfp
description: SFP (Skills for Productivity) is a library of 1,065 agent skills across 24 categories - frontend, backend, mobile, languages, testing, devops, security, agents, agent-dev (skills, plugins, hooks, MCP), AI/ML, AI gateway, data APIs, content, business, industry, networking, research, web3, engineering, memory. Use it first whenever a task might have a matching skill - building UI, APIs or apps, reviewing or securing code, designing agents, plugins or MCP servers, finding a free API or data source, writing or video work - or when the user asks "is there a skill for...", "what can SFP do", or which skill to use.
---

# SFP: pick the right skill

Skills sit in categories. Find the one that fits, read it, and follow it.

For a task with several steps (build, then test, then ship), use the `sfp-hive` skill first: it picks the few skills that fit the task and this project, in order. Use the steps below for a single skill.

## 1. Find the skill

Search `reference/index.md` in this skill's folder. It has one line per skill, grouped by category. Search it for words from the task rather than reading all of it:

- by category: find the `## <category>` heading and read until the next one
- by keyword: search the file, case-insensitive, and try synonyms ("auth", "login", "oauth")

| If the task is about | Category |
|---|---|
| UI, components, motion, accessibility, React, Vue, Next.js | `frontend`, and `design-styles` for a named look |
| APIs, databases, Django, FastAPI, Spring, Laravel, NestJS, Rails, Postgres | `backend` |
| Android, iOS, Flutter, React Native, Swift, Kotlin | `mobile` |
| Go, Rust, Python, C++, C#, Java, Perl idioms | `languages` |
| tests, TDD, E2E, code review, verification | `testing-quality` |
| Docker, Kubernetes, deploys, git, GitHub | `devops` |
| pentesting, vulnerabilities, secure coding | `security` |
| running agents, loops, orchestration, planning, feature pipelines | `agents` |
| building skills, plugins, hooks, MCP servers, agent memory, token budgets | `agent-dev` |
| LLM pipelines, prompts, evals, RAG, PyTorch, ML | `ai-ml` |
| an OmniRoute AI gateway | `ai-gateway` |
| a free API, a data source, internet search or scraping | `data-apis` |
| writing, SEO, video, slides, social posts | `content-media` |
| billing, finance, investors, leads, contracts, email, tickets | `business-ops` |
| healthcare, logistics, customs, inventory, energy | `industry` |
| networks, homelab, Cisco, BGP, VLANs | `networking` |
| research, papers, docs, ADRs, onboarding to a codebase | `research-docs` |
| DeFi, EVM, prediction markets | `web3` |
| architecture, standards, performance, refactoring | `engineering` |
| math proofs, Blender, M5Stack hardware | `science-hardware` |
| job search, CVs, applications | `career` |
| claude-mem, session memory, handoffs, timelines | `memory` |

## 2. Use it

- If the skill is installed, it shows in your skill list: invoke it by name.
- If it is not installed, read its file directly from the SFP checkout and follow it as if it were loaded: `<SFP>/plugins/<category>/skills/<name>/SKILL.md`. Find `<SFP>` in this order:
  1. the folder written in `~/.sfp/root` (the installer writes it),
  2. `~/.claude/plugins/marketplaces/sfp`,
  3. the current project, if it is the SFP repo.
  Skill files point to their own `reference/` and `scripts/` folders by relative path; resolve those against the skill's folder.
- If you cannot find a checkout, tell the user the skill exists and how to get it: `./install.sh install --categories <category>` from the SFP repo, or `/plugin install <category>@sfp` in Claude Code.

## 3. Free APIs and data

For "is there an API for X", "I need data about Y" or "find me a free API", use the `public-apis` skill (category `data-apis`). It searches about 2,100 free APIs and returns links with auth, HTTPS and CORS details. Do not answer from memory when it is available.

## 4. Skills that need a program installed

Some skills guide a program that is installed separately: OmniRoute, FreeLLMAPI, Strix, Agent Reach, claude-mem, the ECC add-on. Each of them carries its own setup, so when the user wants one running, open that skill and follow the note at its top. To see all of them, or to set one up from here:

```bash
python3 scripts/sfp_app.py list
python3 scripts/sfp_app.py status omniroute
python3 scripts/sfp_app.py install omniroute      # shows the plan and changes nothing
python3 scripts/sfp_app.py install omniroute --yes  # only after the user agrees
```

Never run an install without telling the user what it will do. Provider keys and logins are the user's to enter; do not ask for them in chat.

## Rules

- Prefer one specific skill over several vague ones. Most tasks need one or two.
- Load a skill only when the task calls for it. Do not read skills "just in case".
- When two skills overlap (for example `frontend` and `design-styles`), the narrower one wins: use the named style when the user names a style, the general one otherwise.
- If no skill fits, say so and carry on with the task normally.