bouffalolab-skills

BouffaloLab's public collection of agent skills for coding agents (Claude Code, pi, and other SKILL.md-compatible tools).

This repository hosts the skills BouffaloLab publishes for external customers and partners. Each skill lives in its own directory under skills/<name>/ and is defined by a SKILL.md document; agents auto-trigger a skill from its description frontmatter or on explicit request.

Available skills

Skill	Description
wifi-lp-collect	Structured customer information collection for Bouffalo chip WiFi low-power issues. Priority-driven interview (P0/P1/P2, 11 modules), multilingual (中文 / English / 日本語 / 한국어), bilingual Markdown report with completeness check and next steps.

More skills will be published here over time. Watch the repo or follow BouffaloLab for updates.

Installation

Option 1: Claude Code plugin marketplace

# add the marketplace once (from inside Claude Code)
/plugin marketplace add bouffalolab/bouffalolab-skills
# install the plugin
/plugin install bouffalolab-skills@bouffalolab-skills

or non-interactively with the CLI:

claude plugin marketplace add bouffalolab/bouffalolab-skills
claude plugin install bouffalolab-skills@bouffalolab-skills --yes

Option 2: Codex plugin marketplace

codex plugin marketplace add bouffalolab/bouffalolab-skills
codex plugin add bouffalolab-skills@bouffalolab-skills

The marketplace is also visible in the ChatGPT desktop app (Plugins Directory).

Option 3: npx skills

npx skills add bouffalolab/bouffalolab-skills --skill wifi-lp-collect --global --yes

Option 4: Manual copy

Copy the skill's directory into your agent's skills directory, e.g.:

cp -r skills/wifi-lp-collect ~/.claude/skills/
# or
cp -r skills/wifi-lp-collect ~/.agents/skills/

The skill is then auto-discovered — just describe the task to your agent.

Using wifi-lp-collect

Ask your agent in natural language:

I have a customer reporting high sleep current on a BL616. Collect the issue info.

or explicitly:

Use the wifi-lp-collect skill to survey the customer for power consumption debugging.

What happens:

1. The agent greets the customer and asks for the preferred language.
2. Questions are asked module by module, P0 first (basic info, chip config, router, network environment, security, low-power config, diagnostics), then P1 (usage pattern, reproduction), then P2 (comparison data, environment).
3. The customer can reply "不知道 / don't know" at any point or say "完成 / stop / quit" to finish early.
4. A bilingual report is generated and saved as bouffalo-wifi-lp-report-YYYYMMDD-<lang>.md, with completeness markers and next-step recommendations.

Why it helps

• Standardized intake — every issue collected against the same 11-module questionnaire, so no critical data is missed.
• Priority driven — P0 (required) → P1 (important) → P2 (optional); a time-constrained customer can stop after P0 and still get a useful report.
• Friendly by design — accepts "don't know" answers, explains technical terms in plain language, supports early termination at any time.
• Actionable output — completeness check, red flags, and prioritized next steps, ready to hand to your FAE / support team.

Per-skill documentation

• QUICKSTART.md — fastest way to get started
• SKILL.md — the complete workflow
• survey-template.md — every question of the 11 modules
• terminology.md — RSSI, DTIM, Beacon, Keep-Alive, DOZE/SLEEP/DEEP SLEEP, ...
• examples/ — sample reports (full & partial)

Repository structure

bouffalolab-skills/
├── README.md                  # this file
├── AGENTS.md                  # instructions for coding agents working on this repo
├── CLAUDE.md                  # Claude Code specific guidance
├── .claude-plugin/            # Claude Code plugin marketplace
│   ├── marketplace.json       #   marketplace catalog (also readable by Codex, legacy path)
│   └── plugin.json            #   Claude Code plugin manifest
├── .codex-plugin/             # Codex / ChatGPT plugin manifest
│   └── plugin.json
├── .agents/plugins/           # Codex-native marketplace catalog
│   └── marketplace.json
└── skills/
    └── wifi-lp-collect/       # first published skill (more to come)
        ├── SKILL.md
        ├── QUICKSTART.md
        ├── references/        # survey template, terminology
        └── examples/          # sample reports

Contributing

Contributions go through a fork + pull request flow — the main branch is maintainer-only (see branch protection). Start with CONTRIBUTING.md for the full process: fork and branch setup, commit conventions, local verification checklist, and how to submit a PR. Skill authoring conventions live in AGENTS.md.

Support

• Technical support: jsun@bouffalolab.com
• Issues: https://github.com/bouffalolab/bouffalolab-skills/issues

---

Copyright © 2025 Bouffalo Lab Team. All rights reserved.