pedronauck/skills packages 87 agent skills for Claude Code in three buckets

Three buckets with provenance marked on every entry

The repository describes itself plainly as a curated collection of 87 agent skills for Claude Code and compatible AI coding assistants, split into 40 original skills authored in the repository, 34 hand-picked skills including locally authored ones, and 13 marketing, business and writing skills. The catalog is annotated with markers that say which of those three a skill is, and the folder layout matches: skills/mine/, skills/curated/ and skills/marketing/. The origin of the project is stated just as plainly in the title, Pedro Nauck's Skills, and the repository description is My personal favorite skills. That matters for how you evaluate it. This is a curated personal catalogue, not a package with a release cadence, and the value is in the curation and the descriptions rather than in the code.

An agent matches a task to a skill by its description

The mechanism is simpler than the marketing around agent tooling suggests. Skills are structured instruction sets that give an agent expertise in a specific domain. Each skill lives in its own directory under skills/ and contains a SKILL.md file holding metadata, procedures and reference documentation, and the skills follow the agentskills.io specification. What makes the collection usable is how discovery works: skills placed in the ~/.claude/skills/ directory are picked up automatically by Claude Code, and the agent matches a task to a relevant skill based on the description field in each SKILL.md frontmatter. So the description is the entire interface. A well written description decides whether a skill is ever loaded, and that is where the effort in a catalogue like this should be judged.

Install everything, one bucket, or one skill

The recommended path uses the skills CLI to install the repository into your agent skills directory:

npx skills add https://github.com/pedronauck/skills

If you do not want the whole catalogue, the owner/repo/<subpath> shorthand installs a single bucket, and the documented examples target the two buckets you are most likely to want on their own:

# Only the original skills
npx skills add pedronauck/skills/skills/mine

# Only the curated skills
npx skills add pedronauck/skills/skills/curated

Note the doubled path segment in those commands, since the subpath is written relative to the repository root and the skills directory is itself named skills. A single skill can be pinned with the --skill flag:

npx skills add pedronauck/skills/skills/mine --skill react

That granularity is the practical argument for the whole catalogue being a repository rather than a package. Skills are independent directories, so you can take one without taking eighty six others, and adding a bucket later does not mean reinstalling what you already have.

Manual install is a copy or a symlink into ~/.claude/skills

If you would rather not use the CLI, the manual route is to copy or symlink the skills you need into your Claude Code configuration. Copying one skill directory:

# Copy a single skill
cp -r skills/mine/react ~/.claude/skills/react

Symlinking a whole bucket, which keeps the link pointing at your checkout so edits show up immediately:

# Or symlink an entire bucket
ln -s $(pwd)/skills/mine ~/.claude/skills/mine

The symlink form is the one to think about, because it turns the catalogue into a working directory rather than an installed copy. If you plan to adjust a skill to match how your team writes code, symlinking keeps those edits in one place instead of scattering them across the installed copy in your home directory.

The upstream sync is driven by make targets, not by a release

Because the curated bucket contains skills that come from elsewhere, the repository carries machinery for keeping them current. The Makefile exposes four targets beyond help. make update updates verified upstream skills, make check checks upstreams without changing the catalogue, make status shows the offline inventory and local edits, and make test runs the updater tests. Both the update and check targets accept a filter, so a single skill such as firecrawl can be updated on its own, and a whole bucket such as curated can be checked. The test target runs a unittest discovery over the tests/upstream_skills directory, and the scripts/upstream-skills.py script is what actually does the work. The repository also holds skills-lock.json and a .repoclone.rc file, which is where the record of what came from where appears to live.

The catalog names its own overlaps, which saves you time

The skill descriptions are unusually specific, and reading them tells you what you would otherwise discover by trial and error. There is agent-exploration, which dispatches scoped write explorer subagents in parallel for multi-area research, architectural-analysis, a deep architectural audit for dead code, duplication, anti-patterns and code smells, and deep-review, which reviews branch diffs, working trees and pull requests. There is deslop, whose description is to remove AI-generated code slop from the branch diff, naming unnecessary comments, abnormal defensive checks, any casts and deep nesting. The catalogue also flags its own history: golang-master is described as Go 1.21 to 1.26 engineering doctrine with a ten-rule floor and branch-routed references, and it explicitly supersedes the curated golang-pro. That kind of note is worth more than a longer feature list, because it tells you which of the 87 you can safely skip.

No licence, no releases, and a personal scope

The metadata is where this repository is thin, and it is thin in a way that matters for a collection you may copy into your own tooling. The licence for the project is not stated, there are no GitHub releases, and the last push was on 2026-09-14. So there is no version to pin, no changelog to diff, and no answer to the question of what you may redistribute, especially for the curated bucket whose skills were authored elsewhere. The topics of the repository include tooling directories for several agent harnesses, .claude/, .codex/ and .agh/, alongside docs/, scripts/, tests/ and upstream/. The Makefile header dates the copyright to Bernhard Posselt in 2016, which is the boilerplate inherited from the template this project is built on, so the metadata tells you more about the scaffolding than about the skills. Read the licence question before you copy any of this into a commercial product.

Editorial conclusion

Adopt pedronauck/skills if you want a ready-made library of agent instructions to try before writing your own, and if the catalogue descriptions match how your team actually works. Do not adopt it as a production dependency, because the repository states no licence, publishes no releases, and describes itself as a personal collection rather than a maintained product. Before you install, read the description of each skill you plan to enable, check the overlap the catalogue itself flags, such as golang-master superseding an earlier curated skill, and decide whether the upstream sync in scripts/upstream-skills.py is something you want running against your catalogue.

Frequently asked questions

Official sources

1. Issues
2. pedronauck/skills on GitHub
3. README