Zum Inhalt springen
Jefferson Gonçalves
← Zurück zu den Projekten

My Skills

#my-skills

These are Fabio Akita's personal skills — custom-tailored to his machines, subscriptions, and workflow. They are published for reference and inspiration. Treat them as examples of what a skill can look like, then ask your own LLM to write skills for your exact needs. Avoid reusing other people's skills verbatim: they carry someone else's assumptions, paths, and credentials layout.

Single master repository for my agent skills, shared across every AI agent CLI on this machine. Each harness's skills/ directory stays a real directory whose entries are symlinks into this repo — so one edit here reaches all of them.

#Layout

Flat: one directory per skill, each with a SKILL.md.

#Consumers

Harness Skills directory How it links
Claude Code ~/.claude/skills per-skill symlinks
Codex ~/.codex/skills per-skill symlinks
opencode ~/.config/opencode/skills real dir: per-skill symlinks + opencode-only entries
opencode (checkout) ~/opencode/.opencode/skills per-skill symlinks
Kimi ~/.kimi-code/skills whole-dir symlink → this repo
shared ~/.agents/skills per-skill symlinks
alias ~/skills symlink → ~/.claude/skills

Per-skill symlinks rather than replacing each skills/ with one directory symlink: the harness directories also hold entries that must not live here (see below), and per-skill links let those sit alongside untouched.

#Deliberately NOT in this repo

  • omarchy, diagnose-crash — package-owned, symlinked from /usr/share/omarchy/default/agents/skills/. Owned by the omarchy package and replaced on update.
  • oh-my-opencode-slim — opencode-only: the plugin (registered in ~/.claude.json telemetry aside, it runs solely inside opencode) and its config skill live in ~/.config/opencode/skills/ as a real directory, not shared through ~/.claude/skills.
  • assess-team, assess-partners — their own git repos, hosted on a private remote. This repo is public; assessment methodology, templates and reports must not be published here. The seven assess-* skills are nested inside assess-team/skills/.
  • .system — Codex-internal.
  • Vendor bundles: ~/.grok/bundled/skills, ~/.config/crush/anthropic_skills/, ~/.gemini/antigravity-cli/builtin/skills.

#External components (inventory + update procedure)

Third-party skills and plugins installed around this repo. None of them auto-update; re-check occasionally.

Component Source Where it lives Update procedure
humanizer blader/humanizer vendored in this repo (humanizer/) re-pull upstream, strip .git, run python3 humanizer/scripts/validate-package.py
simplify adapted from Addy Osmani's agent-skills this repo (simplify/) one-time adaptation; diff upstream manually when it drifts
codemap, clonedeps oh-my-opencode-slim heritage this repo local forks — no upstream sync
oh-my-opencode-slim npm oh-my-opencode-slim opencode plugin (~/.config/opencode/opencode.json) + config skill in ~/.config/opencode/skills/ opencode plugin oh-my-opencode-slim run from a non-repo dir (it writes local scope otherwise), restart opencode
opencode-openai-codex-auth npm opencode plugin same as above
agent-browser npm agent-browser (mise node global) CLI + SKILL.md vendored in this repo (agent-browser/), symlinked from ~/.config/opencode/skills npm i -g agent-browser@latest, then copy the package's skills/agent-browser/SKILL.md over agent-browser/SKILL.md in this repo
Claude Code plugins (clangd-lsp, frontend-design, rust-analyzer-lsp, typescript-lsp) claude-plugins-official ~/.claude.json claude plugin update <name>
Claude synced skills (skill-creator, docs, pdf, …) claude.ai cloud sync ~/.claude/skills/synced/ managed by account sync
omarchy, diagnose-crash omarchy package symlinks into /usr/share/omarchy/default/agents/skills/ omarchy system updates
Grok / crush / Antigravity bundled skills ship with the tool ~/.grok/bundled/skills, ~/.config/crush/anthropic_skills/, ~/.gemini/antigravity-cli/builtin/skills update the tool

Last checked 2026-09-17 — humanizer 3.0.0 (up to date), oh-my-opencode-slim 2.2.21 stable (beta line 3.0.0-beta.13 available), opencode-openai-codex-auth 4.4.0, agent-browser 0.38.1, all four Claude plugins latest.

#Rules

.gitignore blocks .venv/, __pycache__/, binaries (*.pdf, images, archives), and credential-shaped files (.env, *.key, *token*, *secret*). Assessment PDFs and anything resembling a key or token must never be committed here — the repo is public.

humanizer is vendored from https://github.com/blader/humanizer with its .git stripped. Re-pull upstream manually if it needs updating.

#Adding a skill

mkdir ~/Projects/my-skills/<name>          # add SKILL.md
for h in ~/.claude/skills ~/.agents/skills ~/.codex/skills ~/opencode/.opencode/skills; do
  ln -s ~/Projects/my-skills/<name> "$h/<name>"
done

Harnesses read SKILL.md at startup, so restart a running agent to pick up a newly added skill.

Neue Version verfügbar.