fernando-skills
CLAUDE.md394 starsChanged 6 months agoMITView on GitHub
This repo is a workspace for authoring Claude Code slash-command skills, specifically for a spec-driven development workflow. Skills are written in English, and they instruct the agent to reply in whatever language the user's prompt used. reference/ is a vendored copy of mattpocock/skills kept as a structural/style reference — do not edit it during normal work. Its own reference/CLAUDE.md governs that subtree. Each skill lives under skills/<bucket>/<name>/ (e.g. skills/engineering/spec/) as a directory containing at minimum a SKILL.md. Buckets group skills…
The file
We split this CLAUDE.md into the rules it gives an agent
5 rules. About 0 are common, 0 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 24
Use allowed-tools in frontmatter to restrict what Bash commands a skill may run (see skills/engineering/spec-impl/SKILL.md for an example …
- Very specific: line 26
Use !command
shell snippets insideSKILL.mdto inject live repo state at skill-load time (e.g., current branch, available specs). - Very specific: line 51
Status: after saving is always Draft.
- Very specific: line 60
Read the status line — abort with a standard error message if it does not mean Approved in some language.
- Very specific: line 61
Check the working tree is clean, then create and switch to branch spec-NN-slug.
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Purpose
This repo is a workspace for authoring Claude Code slash-command skills, specifically for a spec-driven development workflow. Skills are written in English, and they instruct the agent to reply in whatever language the user's prompt used.
reference/ is a vendored copy of mattpocock/skills kept as a structural/style reference — do not edit it during normal work. Its own reference/CLAUDE.md governs that subtree.
Skill authoring conventions
Each skill lives under skills/<bucket>/<name>/ (e.g. skills/engineering/spec/) as a directory containing at minimum a SKILL.md. Buckets group skills by domain (engineering/, future: productivity/, misc/, etc.). The YAML frontmatter of SKILL.md must declare:
---
name: skill-name
description: One-line description (in Spanish)
disable-model-invocation: true
argument-hint: [hint]
---Use allowed-tools in frontmatter to restrict what Bash commands a skill may run (see skills/engineering/spec-impl/SKILL.md for an example that limits to read-only git/fs).
Use !command``shell snippets insideSKILL.md to inject live repo state at skill-load time (e.g., current branch, available specs). Embed them at the top of the skill body before the instructions.
Companion files (e.g., template.md) sit next to SKILL.md and are referenced by relative path within the skill body.
The spec workflow
This repo encodes a two-skill pair:
/spec
Guides the user through 4 phases: read project context → clarify with questions (blocks of 3–5) → draft each spec section one at a time with user confirmation → save to specs/NN-slug.md.
On save, it also seeds specs/.spec-config.yml (defaults AutoCreateBranch: true, AutoCommitSteps: false) only if the file is missing — an existing config is never overwritten. This is how the flags that /spec-impl reads get created without the user knowing the file exists.
Output file format: specs/NN-slug.md with a header block:
> **Status:** Draft
> **Depends on:** SPEC 01
> **Date:** YYYY-MM-DD
> **Objective:** One sentence.The date comes from a !date +%F`` snippet injected at skill-load time — never from the model's own sense of the date.
Status after saving is always Draft. State transitions are human-driven — Claude must never change **Status:** automatically.
Valid states: Draft → In review → Approved → Implemented · Obsolete. Equivalents in any language are accepted (Spanish Borrador / En revisión / Aprobado / Implementado / Obsoleto, etc.); /spec-impl matches by meaning, and /spec follows whatever wording the existing specs in the repo already use.
/spec-impl
Accepts <NN-slug> as argument. Phases:
- Locate
specs/<NN-slug>.md. - Read the status line — abort with a standard error message if it does not mean
Approvedin some language. - Check the working tree is clean, then create and switch to branch
spec-NN-slug. If the branch already exists, treat it as a resume and propose a starting step fromgit log. - Implement the spec's plan step-by-step, pausing after each step for diff review. By default it never commits — that stays a human action — unless
AutoCommitStepsistrue(see below). It never pushes.
Branch creation in step 3 is gated by the AutoCreateBranch flag, read at skill-load time from specs/.spec-config.yml via a !cat``snippet. Default (file or value absent) istrue→ branch is created automatically. An explicitfalsemakes the skill ask[y/N] before creating the branch; on decline it implements on the current branch.
Per-step commits in step 4 are gated by the AutoCommitSteps flag from the same file. Default (file or value absent) is false → never commits. An explicit true makes the skill commit each finished step, staging only that step's files, with the bare message spec-NN-step-MM (both numbers zero-padded); review adjustments are folded in with --amend --no-edit.
There is still no runtime config infra — the flags are just values injected into the prompt and interpreted by the model.
At completion, remind the user to verify acceptance criteria and mark the spec Implementado manually before merging.
Distribution
The repo is consumed by users in two ways:
- skills.sh (
npx skills@latest add Klerith/fernando-skills) — auto-discovers public GitHub repos withskills/**/SKILL.md. Just push to GitHub. - Multi-agent installer (
scripts/install-to-agent.sh <agent>) — translates skills for Cursor (.cursor/rules/*.mdc), Codex (AGENTS.mdblock +.codex/skills/), and Antigravity (.antigravity/skills/). Run from the target repo, not this one.
scripts/link-skills.sh symlinks every skill into ~/.claude/skills for local development.
No build or test commands
There is no package manager, build step, or test suite. All skills are plain Markdown files.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

