Skills

Community skills for Pipelit — the self-hosted platform for building and executing LLM agent pipelines through a visual, drag-and-drop interface.

Skills are folders of instructions and resources that Claude loads dynamically to improve performance on specialized tasks. They teach Claude how to complete specific tasks in a repeatable way, whether that's managing development workflows, automating CI/CD pipelines, or orchestrating multi-step processes.

For more information on skills, check out:

• What are skills?
• How to create custom skills
• Equipping agents for the real world with Agent Skills

About This Repository

This repository contains community-built skills designed for use with Pipelit development workflows. Each skill is self-contained in its own folder with a SKILL.md file containing the instructions and metadata that Claude uses.

Skills are organized by category and tracked via skills.json — a package manifest that agents use to compare local skill versions against the repository, similar to how apt or apk manage packages.

Version Management

Skills use a two-tier versioning system, similar to how apt or apk manage packages:

Remote — skills.json

The root skills.json is the remote registry. Agents fetch this single file to discover all available skills, their versions, and checksums.

Local — version.json

Each skill carries its own version.json inside its folder. This is the local manifest that stays with the installed skill. Agents compare their local version.json against the remote skills.json to determine if an update is available.

// <category>/<skill-name>/version.json
{
  "name": "dev-workflow",
  "version": "1.0.0",
  "category": "software_development",
  "description": "Universal dev lifecycle skill...",
  "requires": ["gh", "git", "claude"]
}

Update Flow

Agent fetches remote skills.json
  → compares version against local <skill>/version.json
  → if remote version is newer, downloads and replaces skill folder
  → verifies sha256 checksum after install

When publishing a new skill or updating an existing one, bump the version field in both skills.json and the skill's version.json, then regenerate the sha256 hash:

find skills/<skill-name> -type f | sort | xargs sha256sum | sha256sum | awk '{print $1}'

Available Skills

Software Development

Skill	Version	Description
headless-claude-code	1.1.0	CLI conventions for running Claude Code headlessly — universal invocation template, stream relay
dev-plan	1.0.2	Codebase exploration and structured dev plan generation with draft PR
dev-implement	1.0.1	Implementation from approved dev plans with CI/review triage cycles
opencode-configuration	1.0.0	Configure OpenCode with GLM/MiniMax providers and pre-defined agents
intermediary-delivery	2.0.0	Fire-and-forget outbound message delivery via Telegram scripts

Agent Tools

Skill	Version	Description
agent-browser-init	1.0.0	Install agent-browser standalone binary in Alpine Linux sandbox

Productivity

Skill	Version	Description
meeting-mode	1.0.2	Meeting capture with natural language classification — no prefix triggers needed

headless-claude-code

Conventions and patterns for running Claude Code headlessly via claude -p:

• Universal template — one invocation pattern for all subprocess calls, only timeout and --max-turns vary
• Full tool access — everything except rm -rf, no --model or --allowedTools restrictions
• Stream relay — claude_telegram_relay.py pipes stream-json output to Telegram in real time
• Session management — --resume with --dangerously-skip-permissions re-passed on every resume
• Timeout conventions — single default: 30m timeout, 50 max turns

Requires: claude, python3, TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID environment variables

dev-plan

Codebase exploration and structured development plan generation:

• Explore + plan — claude -p reads the codebase and writes dev-plans/<slug>.md
• Draft PR — plan committed and opened as draft PR for review
• Revision loop — user sends feedback, subprocess revises, loop until approved
• Abort — close PR on user request

Requires: gh (authenticated), git (with push access), headless-claude-code skill

dev-implement

Implementation from approved dev plans with automated review cycles:

• Plan-driven — reads an approved dev-plans/<slug>.md and implements it
• Phased implementation — large plans split into phases automatically when needed
• Review cycle — triages all CI results and PR review comments, fixes issues, loops until green
• Coverage enforcement — never lowers thresholds, writes real tests

Requires: gh (authenticated), git (with push access), headless-claude-code skill

opencode-configuration

Configures OpenCode for use with GLM/MiniMax providers:

• Provider setup — Z.AI (GLM) and MiniMax via environment variables
• Agent definitions — creates plan, implement, review, ci-analysis, coverage agents

Requires: opencode, ZAI_API_KEY and/or MINIMAX_API_KEY environment variables

intermediary-delivery

Fire-and-forget outbound message delivery for orchestrator-level communication:

• Text messages — send.sh sends plain text via Telegram API
• File delivery — send_file.sh sends files with optional captions
• Stateless — scripts deliver and exit, no blocking or polling
• Orchestrator use — status updates between subprocess calls and artifact delivery after exit
• Subprocess visibility — handled by stream relay pattern (see headless-claude-code)

Requires: bash, curl, TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID environment variables

meeting-mode

Real-time working session capture with natural language classification:

• Haiku 4.5 orchestrator — classifies every message naturally (passive/note/action item/research/end)
• No prefix triggers — users just talk, Haiku sorts it out
• Passive capture — zero LLM cost for casual conversation
• Research queries — claude -p spawned only when tool access is needed
• Batch rotation — every 100 messages with Haiku summaries
• End-of-session synthesis — structured summary delivered via intermediary-delivery

Requires: headless-claude-code skill

agent-browser-init

One-time setup for agent-browser in Alpine Linux sandboxes:

• Standalone binary — downloads musl-linked binary from theuselessai/agent-browser-musl releases
• Chromium via apk — installs Chromium as the browser backend
• Architecture detection — supports x86_64 and aarch64
• Environment config — sets AGENT_BROWSER_EXECUTABLE_PATH and agent-browser alias in /etc/profile
• No npm/node — fully standalone, no runtime dependencies

Requires: apk, curl, Alpine Linux sandbox

Installation

Use

Clone the repository:

git clone https://github.com/theuselessai/useless-skills.git

Contribute

Fork, then clone your fork:

gh repo fork theuselessai/useless-skills --clone
cd useless-skills

Creating a Skill

Skills are simple to create — just a folder with a SKILL.md file containing YAML frontmatter and instructions:

---
name: my-skill-name
description: A clear description of what this skill does and when to use it
---

# My Skill Name

[Instructions that Claude will follow when this skill is active]

## Examples
- Example usage 1
- Example usage 2

## Guidelines
- Guideline 1
- Guideline 2

The frontmatter requires only two fields:

• name — A unique identifier for your skill (lowercase, hyphens for spaces)
• description — A complete description of what the skill does and when to use it

Directory Structure

/
├── skills.json                          # Remote package registry
├── README.md
├── CONTRIBUTING.md
├── CLAUDE.md
├── LICENSE
└── skills/
    └── <skill-name>/
        ├── SKILL.md                     # Required — skill definition
        ├── version.json                 # Required — local version manifest
        └── references/                  # Optional — supporting files
            ├── prompts.md
            └── helpers.md

Publishing Checklist

1. Create a folder under the appropriate category (or create a new category)
2. Add a SKILL.md with proper frontmatter and instructions
3. Add a version.json with name, version, category, description, and requires
4. Add a matching entry to the root skills.json
5. Add the category to skills.json → categories if it's new
6. Update the Available Skills table in this README
7. Open a pull request

Contributing

Contributions are welcome! Read CONTRIBUTING.md for the full step-by-step guide — it covers everything from folder structure to PR submission, and is written for both humans and agents.

Skill ideas that would be useful:

• Testing — automated test generation, mutation testing, coverage analysis
• Documentation — API docs, changelogs, architecture decision records
• DevOps — infrastructure as code, deployment pipelines, monitoring setup
• Security — dependency auditing, secret scanning, vulnerability triage

License

Apache 2.0 — see LICENSE for details.