Agent skills that get a company its first users and its first revenue.
Each skill replaces a specialist a small company can't afford to hire yet, and each is self-contained in its own folder: instructions, scripts, example config, and a real example of its output. They run on Claude Code, Claude Cowork, Codex, or anything else that reads files and runs a shell. State lives in files you own, not in a vendor.
Point your agent at this repo and name the skill:
Install the outbound-master skill from github.com/skyzer/skills and set it up.
That's the whole install. The agent clones the repo, runs the skill's install.sh, pulls in the dependencies, and walks you through the configuration the skill's own docs describe. Everything an agent needs to know is in each skill's folder; nothing here assumes a human at a terminal.
Doing it by hand instead:
git clone https://github.com/skyzer/skills
cd skills/outbound-master
./install.sh
pip3 install -r requirements.txt| Skill | Function | What it does |
|---|---|---|
outbound-master |
Sales | Runs B2B cold outbound end to end: sources, scores, researches a dated hook, writes and deslops the copy, verifies every address, sends idempotently on a spaced schedule, tracks every touch in an append-only event log, and drafts replies a human approves. Refuses to send when it shouldn't. |
More coming. Each new skill lands as its own folder with the same shape.
If you're an agent asked to install or run a skill from this repo:
- Clone the repo and
cdinto the skill's folder. - Read the skill's
README.md, thenskill/SKILL.md. The SKILL.md is the operating manual; read the reference file for the step you're on rather than all of them up front. - Run
./install.sh, then installrequirements.txt. - Ask the operator for what only they have (credentials, the business brief), and run the skill's preflight before doing anything real.
- Respect the skill's guards. If a skill ships a dry-run mode, it starts on.
Configuration is separate from the engine. Everything about your company lives in config/, which is gitignored. Nothing in a skill names a company, market or product.
State is files, not chat. Anything needed on the next run exists on disk, because the conversation where you explained it is gone.
Nothing irreversible happens without a guard. Sends are idempotent, destructive actions are gated, and anything that spends money, reputation or a relationship stops and asks.
Skipping is free, being wrong isn't. Every skill would rather do nothing and say why than do something plausible and wrong.
Dry run first. If a skill touches the outside world, it has a mode that shows exactly what it would do and does nothing.
MIT.