# Skills

Professional skills for nasty bugs, hard decisions and work that needs to be right.

Most public skills help an agent perform one narrow operation. These are for professional work where the brief is incomplete, the system is messy, or a polished answer can still be wrong.

I like [Matt Pocock's skills](https://github.com/mattpocock/skills) and his warning against cognitive surrender. I agree that the human should keep judgment. I simply put more of the legwork on AI. There is no point asking a human to repeat known best practices, collect routine evidence or catch basic mistakes that an agent can handle.

Factories are my answer. Give AI enough context, authority, professional practice and verification to do more of the work well. I question pre-AI software development, keep the principles that still serve a purpose, and rebuild the workflow around what AI can now do.

My approach is holistic. I combine experience in software development, product, privacy, security and work organization. The public skills contain the reusable layer. More specialized agents and factories remain private.

## Choose the right one

- The cause is unknown or the bug keeps coming back: use Dig Deeper.
- Diagnosis needs tools, reproduction or controlled experiments: use Dig Deeper Probe.
- The request may hide other intents, domains or unknown unknowns: use Smart Shot.
- Your prompt is weak and a moderately better result is enough: use Fast Shot.
- A bounded task needs basic mistakes removed before delivery: use SFLO Quick.
- A skill has become expensive to load: use Skill Compressor.

## Dig Deeper

[Dig Deeper](/skills/dig-deeper/) is great for nasty, deep bug diagnosis. It uses logs, system state, documentation, code and other available evidence. Probe mode can install tools, reproduce failures and run controlled experiments when observation is not enough.

## Smart Shot

[Smart Shot](/skills/smart-shot/) discovers hidden intents, affected domains and unknown unknowns. It assembles ad hoc expert agents from the work itself, uses the right tools, acts within authority and independently tests the result.

## Fast Shot

[Fast Shot](/skills/fast-shot/) is a lighter, much faster Smart Shot. It gets a moderately better result from a short or underspecified prompt without paying for a specialist team.

## SFLO Quick

[SFLO Quick](/skills/sflo-quick/) executes any bounded task through a lightweight plan-do-check-act cycle. It catches basic mistakes before delivery without the weight of a full factory.

## Skill Compressor

[Skill Compressor](/skills/skill-compressor/) saves tokens and improves speed only when tests show that the skill still activates and behaves correctly.

## Why these are different

- Dig Deeper uses independent Troubleshooter and Interrogator roles to stop a plausible theory from becoming a fake root cause.

- Smart Shot discovers the work before staffing it. Its expert roster is created for the request, not picked from a fixed list.

- Fast Shot is for the times when I am too lazy to write a proper prompt and want a moderately better result without paying for the full team.

- SFLO Quick removes the basic mistakes AI often leaves behind. It is intentionally lightweight for demos, small projects and work where speed matters more than TDD, security testing or a full factory. Regular SFLO remains a software factory.

- Skill Compressor is a measured utility. Shorter wins only when routing and behaviour still pass.

## Install

Install the collection for Codex and other compatible agents:

```shell
npx skills@latest add simonasrazm/skills
```

Claude Code can install the repository as a plugin marketplace:

```shell
claude plugin marketplace add simonasrazm/skills
claude plugin install simon-skills@simonasrazm
```

The [source repository](https://github.com/simonasrazm/skills) is MIT licensed. Each skill page links to its source and current skills.sh listing. Agents can use the [machine catalogue](/skills/index.json) for stable descriptions, documentation, source and install fields.

Source: https://finallydone.ai/skills/