swyx skills
A personal collection of about 130 Claude Code and Codex skills: media pipelines, Cloudflare, YouTube, design and task review
- 198 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Some skills drive a browser, publish videos to YouTube and work with API credentials
- Skills download media and run scripts from the cloned repository
Install
- Works in
Manual install
git clone https://github.com/swyxio/skills.git ~/Work/skills && mkdir -p ~/.agents/skills && ln -s ~/Work/skills/review-thread ~/.agents/skills/review-threadInstalls one skill by symlink. Repeat ln -s for the others. Cursor also reads ~/.agents/skills/.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Some skills drive a browser, publish videos to YouTube and work with API credentials
- Skills download media and run scripts from the cloned repository
This is third-party code. Review the repository files before installing.
What it does
Each repository folder holds one skill with a `SKILL.md`: a trigger description and the workflow the agent reads when the skill applies. Scripts, reference files and assets sit alongside. A routing table in the README says which skill to start with and which one takes the next stage. Skill groups. Media: `media-transform`, `download-video`, `transcribe-anything`, `thumbnail-extraction`, `annotated-video-production` and YouTube skills for the API and for Studio through Chrome. Cloud: `cloudflare`, `durable-objects`, `wrangler`, `workers-best-practices`, `agents-sdk`, `sandbox-sdk`. Design and checks: `design-apps-with-imagegen`, `visual-playtest`. Process: `review-thread`, `ceo-creator`, `cassandra-creator`, `security-hardening`, `release-readiness-hardening`. Conference skills for Accelevents are included too.
Who it is for. People who build media pipelines and Cloudflare Workers apps and want ready workflows from an experienced author to study or reuse.
Good fit when
- A video has to be downloaded, transcribed, given thumbnails and published to YouTube in one pipeline
- A Cloudflare Workers app needs correct handling of Durable Objects, Wrangler and the Agents SDK
- A site or app has to be inspected as a user would and layout defects found on different screens
- Midway through a long task you need an honest assessment of progress, what remains and agent run time
- You want a ready skill as a pattern for structure: trigger description, references, scripts
Not a fit when
- You need one narrow skill with a vetted install: this is a personal working collection and some skills assume the author's projects and credentials
- Writing skills tuned to the author's style: `swyx-writing` and `tweet-writing` fit one specific person and will give someone else's text his manner
- You have no YouTube, Cloudflare or Airtable credentials: the skills that depend on them will not run
Example request
Use `review-thread`: assess how much is left in the migration, what is verified versus merely written, and which three tasks are worth running in parallelLimitations
- The author calls the repository a personal working collection: some skills assume his projects, preferences, credentials and installed tools. Read each skill's prerequisites before running it.
- Installing places instructions only. Workflow dependencies and access to APIs, browsers and publishing are yours to set up.
- Skills drive Chrome through computer use, download videos and publish to YouTube: actions on external services are real.
- Some skills are tuned to the author's writing manner.
How to disable. Delete the symlink of the skill from `~/.agents/skills/` or `~/.claude/skills/`, for example `rm ~/.agents/skills/review-thread`. The repository clone can stay or go.
FAQ
How do I install the swyx skills in Claude Code?
Clone the repository and link the skill folder into `~/.claude/skills/`, for example `ln -s ~/Work/skills/review-thread ~/.claude/skills/review-thread`. The skill is called as `/review-thread`.
Can I install all the skills at once?
The author advises linking whole skill folders one by one from a single clone so scripts and references stay together. Repeat the link command for each skill you need.
Which skill should I start with?
`review-thread` fits a progress review. For video pipelines start with `media-transform`, for an app redesign with `design-apps-with-imagegen`. The routing table is in the README.
Do I need keys and access?
For some skills yes: the YouTube API, Cloudflare, Airtable and browser accounts. Installing grants no access, and each `SKILL.md` lists its requirements.
How do I validate my own skill against the repository pattern?
The author keeps a metadata validator: `uv run --locked python scripts/validate_skill.py <skill>` after `uv sync --locked`.
Discussion
No questions or reviews yet.
Related
Skills11 skills for Claude Code: SKILL.md authoring, prompt tuning, refactoring, change review and planning in files.
A plugin of seven skills for Claude Code and Codex: prompts, subagents, skill authoring, editing, web search and a browser.
A working folder for Claude Code where the agent asks questions and writes a new skill by Anthropic's guidance, not from memory.
A catalog of nearly two hundred skills: code delivery, review, debugging, AI governance, product work and agent coordination.
The xskill CLI gathers sanitized session trajectories from Claude Code, Codex and OpenCode and grows a shared skill library.
An open catalogue of over three thousand skills, each backed by a real project, with industry collections and categories.
README in short
It opens with a one-skill install example that checks the link is readable and a folder table for Codex, Claude Code and Cursor. A routing table by task follows, with orchestrators and atomic skills. It ends with acceptance principles: what counts as blocking, how not to expand scope and how to validate skills with the validator.
README badge
Original SKILL.md
--- name: review-thread description: Review task progress, goal alignment, agent-time ETA, and worthwhile parallel work. Use when the user invokes /review-thread or requests a progress or scope assessment. --- # Review thread Give a concise, candid assessment of progress toward the user's intended outcome. Keep reporting lightweight and continue authorized work. ## 1. Progress and direction Infer what success means to the user; state consequential assumptions. Explain what now works, what remains, and what is verified. Distinguish implementation, passing tests, and actual user-visible success. Assess whether we are fixing the right problem: - What can we remove without weakening the outcome? - What must we add to fully satisfy the use case? Use independent judgment: do not invent scope problems to agree with the user. Recommend concrete corrections when warranted, and correct earlier claims plainly. Report outcomes rather than tool-call chronology. ## 2. Next steps and ETA Name the completion endpoint. Estimate AI-agent wall-clock time using observed execution speed and tool runtimes. Account for available parallelism, dependencies, shared resources, and coordination. Use the longest dependent sequence rather than adding all work together. Calibration anchors: - Three independent two-minute checks running concurrently take roughly two minutes plus coordination. - A suite observed to run in 20 seconds gets a seconds-scale allowance. - Narrower follow-ups usually take less time than completed broader work unless a new dependency changes that. - Approval, credentials, and human replies are external waits; report them separately. Give a likely ETA and conditional upper bound with key assumptions. Avoid generic padding or forced optimism. If calibration is weak, estimate the next informative checkpoint. Revise materially changed estimates using actual elapsed time. ## 3. Parallel opportunities When worthwhile, suggest a numbered list of up to three independent tasks, ranked by value. State each task's expected result, why it helps now, and any dependency or conflict.