solo-skills: 38 Claude Code agent skills for one-person businesses

What solo-skills is, and the solopreneur it is written for

The README opens with a claim rather than a feature list: the author automated 58 jobs while running a community called GET100, now in its sixth cohort, and published 38 of those procedures here. The audience is narrow and stated plainly. It is one person doing product videos, ebook PDFs, blog posts, meeting notes, customer messages, and reply threads without staff. The repository does not try to be a general automation framework. It is a shelf of procedures that one operator extracted from a working setup.

That origin shapes what is inside. Each skill is described as a procedure document an agent reads and performs, not a library you import. Some skills ship the code as well, which the README counts as 10 skills and 19 files. The rest are described as tied to the author's accounts and servers, with only the procedure and the pitfalls written down. fleet.md holds the full list of 58, including the ones that were never published.

The README also documents what was deliberately withheld. Four automations stay private because a business registration number, a bank account, or a client's real name is embedded in the procedure, and removing it stops the skill from running. Third-party names and Notion and Discord identifiers were replaced with placeholders in the published set. That is an unusual level of disclosure about a gap, and it tells you the published 38 are the subset that survived a scrubbing pass.

How the skills are structured, and what the safety notice covers

The format is the SKILL.md convention. The README states the skills run as-is in Claude Code, Codex, OpenCode, and other agents that read that convention, which means the repository has no runtime of its own. There is no daemon, no server to start, and no dependency graph to resolve. The agent reads the procedure and executes it with whatever tools it already has.

Twenty-two of the 38 skills carry a notice at the top titled, in the original, what this skill does to your system. The README says it tells you which permissions the skill needs, what it actually publishes or sends, and whether the action can be undone. Read that notice before installing, because the answer differs sharply across the set. threads-reply is built to default to a dry run and only publishes with an explicit --go flag. naver-mail sends through Naver SMTP with attachment support, which is not reversible once the message leaves. remote-offload is documented to stop rather than fall back to local execution when the remote machine is down, a design choice that prevents silent load on the wrong host.

The scripts are not uniform in their assumptions. claude-codex-fallback runs Claude first and re-runs on Codex only when a limit is exceeded. discord-reminder registers one-off reminders with launchd, which is macOS-specific, and deletes them after sending. The lecture-recording-captions skill splits into nine scripts across transcription, proper-noun correction, caption assembly, and compositing, so a wrong step can be re-run alone. That split is the most interesting engineering decision in the repository: the README says trying to do it in one pass meant starting over whenever something failed midway.

Installing solo-skills and running your first skill

The README gives three installation routes and marks the plugin route as the recommended one. The two commands register the repository as a marketplace and then install the whole set. Run them inside an agent that supports the plugin marketplace commands.

/plugin marketplace add bam-bam-2/solo-skills
/plugin install solo-skills@bam-bam-2

If you want one skill instead of 38, the README shows the npx route with a skill name. The example uses naver-blog-post, which is the name as it appears in the README.

npx skills add bam-bam-2/solo-skills --skill naver-blog-post

The third route is a manual copy into the Claude skills directory. This assumes the default path and that the directory exists.

git clone https://github.com/bam-bam-2/solo-skills.git
cp -R solo-skills/skills/* ~/.claude/skills/

For a first real run, pick threads-reply. It is one of the skills that ships a script, publish-thread.mjs, and the README states its default is a dry run. You see the composed thread without anything being posted, and only the --go flag publishes. That makes it the cheapest way to confirm the skill loaded and your agent can execute the procedure. Before you run it, open the notice at the top of the skill folder and check the permission list. The README does not document a rollback for skills that publish or send, so the dry run is the only safe rehearsal the repository offers.

Where solo-skills breaks: same-origin, macOS, and Naver assumptions

The web-demo-video skill is the clearest example of a constraint that will bite. It builds demo videos without screen recording: a stage page embeds the target site in an iframe inside a mock browser window, a fake cursor SVG fires real click events into the iframe, and a single window.__tick function renders each frame deterministically so repeated captures produce the same video. The README states the stage page and the target site must be on the same origin. Different ports block iframe.contentDocument access and the whole manipulation stops working. The stage file is therefore placed temporarily inside the project folder and removed afterwards, because leaving it there means it ships with the next deploy. That is a real operational hazard, not a footnote.

The platform assumptions are just as concrete. discord-reminder registers jobs with launchd, so it is macOS-only. naver-mail depends on Naver SMTP, so it presumes a Naver account. The README says the entire 58-skill fleet runs on a single Mac mini, which explains both choices and also tells you the published set inherits that environment. If you run Linux or Windows, several skills need the scheduling layer rewritten before they do anything.

book-pdf carries a layout trap the README calls out directly. Using flexbox for part and chapter openers breaks Paged.js page-break calculation. There is also a separate rule preventing paragraphs from splitting awkwardly across pages. These are the kind of details that only surface after a rendered book, and they are documented rather than hidden.

Finally, there are no releases. The repository has no version tags, so an update means pulling the branch again. The README's own update notes describe a 26-to-38 expansion after an 11-day gap, and a blog skill split into three because one set of title, length, and tone rules could not serve informational posts and review posts at once. You are tracking a moving branch, not a pinned artifact.

solo-skills against a general-purpose agent framework

The obvious comparison is LangChain or a similar agent framework. The difference is in what gets stored. A framework gives you abstractions for tools, memory, and chains, and you write the behaviour. solo-skills gives you the behaviour already written as prose procedures, with a handful of scripts attached, and no abstraction layer at all. There is nothing to compose programmatically. You install a skill folder and an agent reads it.

That trade runs in both directions. You get a procedure that already encodes failures the author hit, such as the same-origin requirement in web-demo-video or the proper-noun dictionary that must be loaded before transcription in lecture-recording-captions. The README states that skipping the dictionary means personal names and brand names land as wrong words and the whole subtitle track needs rework. You do not get versioning, tests, or a support channel. A framework with a test suite will tell you when an upgrade breaks you. This repository will not.

The second comparison is to Anthropic's own published skills. The README says skill-creator and slack-gif-creator were excluded because they are not the author's work. So solo-skills is not a superset of the official set. It sits beside it, and the overlap is zero by the author's own accounting.

Where the repository does something a framework would not: threads-long-thread derives its format from 3,870 public posts scraped from two accounts that perform well, and records per-post character counts and line-break positions. It also ships ai_smell.mjs, which scores a draft by how far its line-length distribution deviates from the measured one, rather than by a subjective read. That is a measurable criterion, and it is the kind of thing you would otherwise have to build.

Licence, maintenance, and what each update costs you

The licence is MIT, declared in the README badge and present as a LICENSE file at the repository root. MIT permits commercial use and modification, and requires the copyright notice and permission notice to be retained. That matters for the withheld four skills: they are not in the repository, so no licence question arises for them. For the published set, if you fork a skill and ship it inside a product, keep the notice. This is a description of the licence text, not legal advice; read LICENSE yourself if the distinction matters to your situation.

Maintenance is visible and recent. The last push to the default branch was on 2026-09-14, three days before this writing, so the repository is not dormant. It is not archived. There are no tagged releases, which means there is no changelog to diff and no way to pin a known-good state. The README's update notes are the closest thing to a changelog, and they are written as narrative rather than as a version list. The 26-to-38 expansion is described there, along with 12 existing skills revised to reflect what broke.

Upgrade cost is therefore manual. You re-clone or re-pull, then diff the skill folders you actually use against your local copies, because any edits you made will be overwritten by a copy command like the one in the install section. If you customized a skill, keep your version outside the folder that the install command writes to. The repository does not document a merge path.

Editorial conclusion

Adopt solo-skills if you already run Claude Code, Codex, or OpenCode and you want a working procedure for a job you repeat, such as turning a manuscript into a Paged.js PDF or cutting a lecture recording into captioned video. Skip it if you need a supported library with a version history: there are no releases, the README says the remaining 20 skills stay private, and several scripts assume macOS launchd or a Naver account. Before installing anything, read the notice at the top of the skill you picked and check whether it publishes or sends on your behalf, then try threads-reply, whose default is a dry run and which only publishes with --go.

Frequently asked questions

Official sources

1. bam-bam-2/solo-skills on GitHub
2. Issues
3. License: MIT
4. Project website
5. README