# Andrej Hide Skills

Open-source skills inspired by Andrej Karpathy's notes on agentic coding (Dec 2025). The focus is practical: reduce wrong assumptions, avoid bloat, keep plans light, and validate with tests and conceptual review.

## Quick start

Install:

```bash
npx skills add Hideturno/Andrej_Hide_Skills
```

Use:

```
/AndrejHide
```

## Inspiration and evolution

This repo started directly from Andrej's post. After drafting the skills, I discovered another repo that packaged similar ideas into a single CLAUDE.md (see below). I used it as a benchmark and pushed this repo further by splitting the principles into reusable skills, adding explicit workflows, and making the set installable via skills.sh.

## Whats inside

- `skills/`: individual skills, each with a `SKILL.md`.
- `SKILLS.md`: short index of skills.
- `WORKFLOWS.md`: the standard workflows that chain skills.

## Skills

- `agent-coding-workflows`: select and run the standard workflows.
- `assumption-clarifier`: surface missing info and make assumptions explicit.
- `plan-lite`: short plan with risks and verification.
- `minimal-diff-implementer`: smallest safe change, no unnecessary refactors.
- `criteria-test-loop`: success criteria + tests, naive first then optimize.
- `conceptual-reviewer`: catch subtle logic errors and overcomplication.

## Workflows

Use the `agent-coding-workflows` skill to choose a workflow:

- Standard change
- Algo or performance
- Urgent bugfix

## Agent usage (slash)

You can call the workflow agent directly:

```
/flow standard
/flow algo
/flow bugfix
/AndrejHide
```

## Override thresholds (optional)

You can override the default thresholds per task:

```
thresholds: files=5 lines=300 minutes=45
```

Example with auto workflow:

```
/AndrejHide
thresholds: files=5 lines=300 minutes=45
```

## Principles to skills map

- Think Before Coding -> `assumption-clarifier`, `plan-lite`
- Simplicity First -> `minimal-diff-implementer`
- Surgical Changes -> `minimal-diff-implementer`
- Goal-Driven Execution -> `criteria-test-loop`, `agent-coding-workflows`

## Why this repo is different (and better for daily use)

- Modular skills instead of a single file, so you can apply only what you need.
- A workflow skill that chains principles in the right order for real tasks.
- Explicit success-criteria loop (tests first, naive correct, then optimize).
- A dedicated conceptual review pass to catch subtle logic errors.
- Compatible with skills.sh for one-line install and reuse across projects.

## Related repo

I found this after drafting my skills and used it to improve the set:

- `forrestchang/andrej-karpathy-skills`

## Install (skills.sh)

Use the Skills CLI:

```bash
npx skills add Hideturno/Andrej_Hide_Skills
```

## Install (CLAUDE.md)

New project:

```bash
curl -o CLAUDE.md https://raw.githubusercontent.com/Hideturno/Andrej_Hide_Skills/main/CLAUDE.md
```

Existing project (append):

```bash
echo "" >> CLAUDE.md
curl https://raw.githubusercontent.com/Hideturno/Andrej_Hide_Skills/main/CLAUDE.md >> CLAUDE.md
```

## Install (.claude/skills)

```bash
mkdir -p .claude/skills
curl -o .claude/skills/Andrej_hideturno_guideline.md https://raw.githubusercontent.com/Hideturno/Andrej_Hide_Skills/main/.claude/skills/Andrej_hideturno_guideline.md
```

## License

MIT. See `LICENSE`.