RPA Skills

Pavel Rykov's agent-skills marketplace — a plugin catalog for AI agents in Claude Code, OpenAI Codex, Cursor, Kimi Code CLI, and so on. Connect the marketplace with one command and install any skill from it, or install every skill on its own.

Motivation: notes on vibe coding and the prompt collection cursor-vibe-prompts are packaged as skills so you do not paste the same long instructions into chat every time.

This repository is the marketplace itself, not the skills. Each skill lives in its own repository so it can be versioned, installed, and depended on independently. This repo keeps the catalog in two marketplace manifests — .claude-plugin/marketplace.json for Claude Code and .agents/plugins/marketplace.json for Codex — plus an install.sh for folder-based agents like Cursor, the authoring guide (AGENTS.md), and the links below.

Skills

Skill	Purpose
/rpa-init	Warm up context on the repo: study code, docs and test code, set up the dev environment, run tests, write a short project report. No extra brief.
/rpa-gen-rules	Create or refresh agent rules for Cursor (.cursor/rules/*.mdc), Claude Code (.claude/), and Codex (.codex/ hook bridge that auto-attaches Cursor rules by glob). Layered-cake architecture + BDD, with a mandatory Rules Sync step across agent trees. Rules and AGENTS.md are written in English unless you ask otherwise.
/rpa-feat	Add a feature strictly by BDD: plan, tests (red), implementation (green), full run, docs and examples, linter at the end. Requires a clear task.
/rpa-bugfix	Reproduction test first, then fix, full test run, short report. Requires a clear bug description.
/logika	Classical formal logic from G. Chelpanov's textbook (concepts, judgments, syllogisms, induction, fallacies).
/token-cost	Estimate the floor cost of a self-hosted LLM token (electricity + hardware amortization), not the market price.
/mikrotik-config-gen	Generate secure MikroTik RouterOS v7 configs from plain language, or review and harden an existing /export. Hardening baseline, four presets, per-model device reference, offline validator.
/screenshotting-gui	Capture clean, repeatable screenshots of a GUI desktop app (IDEs, browsers, native apps) on an isolated Xvfb display, driven with python-Xlib (XTEST), annotated with ImageMagick (highlight box, crop, arrow). Never touches the operator's live desktop.
/gpu-server-setup	Prepare a Linux server with NVIDIA GPUs for LLM workloads: driver + CUDA from the official repo, Docker + NVIDIA Container Toolkit as the mandatory base, staged verification gates, optional compose presets for vLLM, Infinity, OpenWebUI, Ollama and llama.cpp, plus GPU power-limit setup persisted via systemd.
/rdp-agent	Drive a remote Windows machine over RDP: the RDP client runs on an isolated Xvfb display, the operator watches the same screen in a VNC window and can take over by hand, the session survives between agent turns, and a lock screen is recovered by an NLA reconnect instead of typing the password.

Input for /rpa-* — /rpa-init and /rpa-gen-rules need no extra briefing. /rpa-feat and /rpa-bugfix need you to pass what to do (for example issue text), otherwise they will not work correctly.

Each skill repository has its own README.md with usage, install, and attribution.

Install

This catalog ships two marketplace manifests so it can be connected as a marketplace from more than one agent, plus a plain installer script for agents that have no remote marketplace at all:

File	Consumed by	How you connect it
.claude-plugin/marketplace.json	Claude Code (also read by Codex as a legacy fallback)	/plugin marketplace add EvilFreelancer/rpa-skills
.agents/plugins/marketplace.json	Codex	codex plugin marketplace add EvilFreelancer/rpa-skills
install.sh	Cursor and any folder-based agent	clones skills into ~/.agents/skills/

All three point at the per-skill repositories listed above. Make sure those repositories exist and are pushed (they are split out from this one), then the entries resolve.

Claude Code

Add this repo as a plugin marketplace, then install any skill from it:

/plugin marketplace add EvilFreelancer/rpa-skills
/plugin install rpa-init@rpa-skills
/plugin install logika@rpa-skills
# ...and so on

Codex

Codex has its own repo marketplace. Add this catalog, then browse/install from /plugins:

codex plugin marketplace add EvilFreelancer/rpa-skills
codex plugin marketplace list        # verify it resolved
# then in the CLI: run `codex`, open `/plugins`, pick the "RPA Skills" tab, install what you need

Codex reads .agents/plugins/marketplace.json. It can also read the Claude manifest above as a legacy-compatible source, so either entry point works.

Cursor (and other folder-based agents)

Cursor has no remote plugin marketplace — it discovers skills only from directories (.agents/skills/, .cursor/skills/, and their ~/ globals). So instead of "connecting" the catalog, you install the skill folders into a skills root the agent reads:

# install/update the whole catalog into ~/.agents/skills (read by Cursor and Codex)
curl -fsSL https://raw.githubusercontent.com/EvilFreelancer/rpa-skills/main/install.sh | bash

# or clone the repo and run it, optionally targeting a specific root / subset
git clone https://github.com/EvilFreelancer/rpa-skills && cd rpa-skills
./install.sh                       # -> ~/.agents/skills
./install.sh -d ~/.cursor/skills   # -> a Cursor-specific root
./install.sh logika rpa-init       # only the named skills

Then reload the Cursor window (Cmd/Ctrl+Shift+P → Developer: Reload Window).

A single skill, by hand

Every skill repo keeps its SKILL.md at the repo root, so you can clone one straight into a skills root. The directory name must match the name field in SKILL.md:

git clone https://github.com/EvilFreelancer/logika ~/.agents/skills/logika

Typical skill layout

<skill>/
├─ .claude-plugin/
│  └─ plugin.json        # plugin manifest (name, version, description) — no skills field
├─ SKILL.md              # YAML frontmatter + agent instructions (canonical, loaded from the repo root)
├─ references/ | scripts/   # optional bundled assets, referenced relative to SKILL.md
├─ README.md             # human overview
└─ LICENSE

A plugin with SKILL.md at the repo root, no skills/ directory, and no skills field in plugin.json is loaded by Claude Code (v2.1.142+) as a single-skill plugin; the invocation name comes from the name field in SKILL.md. That name should match the slash command — the rpa-* skills use hyphenated names such as /rpa-init.

Making the agent use a skill

1. Slash command — type / in agent chat and pick the skill by its name, or type it directly, e.g. /logika or /rpa-init.
2. @ context — type @ and attach the skill (or its folder / SKILL.md) to ground the message.
3. Automatic selection — the agent may load a skill on its own when your request matches the description in SKILL.md, so clear trigger wording in that field helps.

Maintaining the catalog

See AGENTS.md for the collection-wide process: how the two marketplace manifests, the Skills table, and install.sh mirror each skill's metadata, and the sync rule to follow when a skill bumps its version, description, or tags. For authoring an individual skill, see the AGENTS.md in that skill's own repository.

License

MIT — see LICENSE. Each split-out skill repository carries its own copy of the same license.