[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-81131":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":15,"stars7d":17,"stars30d":18,"stars90d":16,"forks30d":16,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":24,"hasPages":22,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":26,"readmeContent":27,"aiSummary":28,"trendingCount":16,"starSnapshotCount":16,"syncStatus":14,"lastSyncTime":29,"discoverSource":30},81131,"show-us-your-agent-skills","hugobowne\u002Fshow-us-your-agent-skills","hugobowne","Companion repo to our livestream series Show Us Your (Agent) Skills","https:\u002F\u002Fhugobowne.github.io\u002Fagent-skills",null,"JavaScript",39,4,2,1,0,5,7,3,2.1,"MIT License",false,"main",true,[],"2026-06-12 02:04:11","![Show Us Your Agent Skills](images\u002Fskillz-1.jpg)\n\nWhat are people at the top of the game building with AI agents, and how are\nthey doing it?\n\nAre they Claudemaxxing with 8 terminals open at once? Or adversarially\ntesting Opus 4.7 generated code with OpenAI Codex? Do they define suites\nand swarms of sub-agents, or use AGENTS.md and agent skills?\n\nWhat do they love about building with agents? What do they hate? What tips\nand tricks do they use to supercharge their workflows?\n\nThomas Wiecki (PyMC Labs) and Hugo Bowne-Anderson (Vanishing Gradients) are\non a mission to find out. Think Excel World Championships meets Eurovision.\n\nThis repo turns each episode into browsable, forkable artifacts: markdown\nskills, workflow writeups, and tool references you can read, copy, and\nadapt to your own setup.\n\n## Installation (npx skills)\n\nYou can install the skills in this repo with:\n\n```\nnpx skills add https:\u002F\u002Fgithub.com\u002Fhugobowne\u002Fshow-us-your-agent-skills\n```\n\nInstall one skill only:\n\n```\nnpx skills add https:\u002F\u002Fgithub.com\u002Fhugobowne\u002Fshow-us-your-agent-skills --skill explain\n```\n\nCheck for updates:\n\n```\nnpx skills check\nnpx skills update\n```\n\nThese skills are snapshots from the corresponding live streams. Creators often iterate on their own versions afterwards, so check each skill's README for instructions to pull the latest from the creator when one is linked.\n\n## Episode 1: Wes McKinney, Jeremiah Lowin, Randy Olson\n\n[Watch on YouTube](https:\u002F\u002Fyoutube.com\u002Flive\u002FPq3xuChdwxQ)\n\n| Skill | What it does | Guest | Watch |\n|-------|--------------|-------|-------|\n| [explain](skills\u002Fexplain) | Agent narrates what it just did, like a teammate handing off. | Jeremiah Lowin (Prefect, FastMCP) | [00:46:14](https:\u002F\u002Fyoutube.com\u002Flive\u002FPq3xuChdwxQ?t=2774) |\n| [github-reply](skills\u002Fgithub-reply) | Replies to GitHub contributors in your voice, no \"Great work, but rejected\" sandwiches. | Jeremiah Lowin (Prefect, FastMCP) | [00:54:08](https:\u002F\u002Fyoutube.com\u002Flive\u002FPq3xuChdwxQ?t=3248) |\n| [ship-it](skills\u002Fship-it) | Re-trains \"ship it\" to mean *open a PR*, not merge. | Jeremiah Lowin (Prefect, FastMCP) | [00:54:52](https:\u002F\u002Fyoutube.com\u002Flive\u002FPq3xuChdwxQ?t=3292) |\n| [high-signal-chart-workflow](skills\u002Fhigh-signal-chart-workflow) | Turns a one-line idea into a Tufte-style chart, with an LLM-as-judge verifier loop. | Randy Olson (Goodeye Labs, r\u002Fdataisbeautiful) | [01:12:37](https:\u002F\u002Fyoutube.com\u002Flive\u002FPq3xuChdwxQ?t=4357) |\n| [8-bit-video-gen](skills\u002F8-bit-video-gen) | Turns guest headshots into short 8-bit pixel-art video clips for livestream intros and cutaways. | Show Us Your Agent Skills | [Episode 1](https:\u002F\u002Fyoutube.com\u002Flive\u002FPq3xuChdwxQ) |\n\nWorkflow writeups are coming next, starting with Wes McKinney's stack:\n\n- **Agents reviewing agents.** A daemon reads every commit your agents make, so by the time a PR merges, code has been read by agents 4–5 times.\n- **A fleet of long-running sessions.** 4–5 Superpowers projects spec'd and implementing in parallel, unattended; one plan ran 14 hours and 45 tasks without him touching it.\n- **\"Off the rails?\" review.** No line-level reading. The only question Wes asks is whether the agent strayed structurally or chased scope creep.\n\n## Episode 2: Hilary Mason, Bryan Bischof, Eric Ma, Tomasz Tunguz\n\n[Watch on YouTube](https:\u002F\u002Fyoutube.com\u002Flive\u002Fl37PR-OkYKA)\n\n| Skill | What it does | Guest | Watch |\n|-------|--------------|-------|-------|\n| [prompt-refinement](skills\u002Fprompt-refinement) | Interview the user's intent, ask for three variations at different magnitudes of change, score against a rubric you wrote up front. | Hilary Mason (Hidden Door) | [01:01:00](https:\u002F\u002Fyoutube.com\u002Flive\u002Fl37PR-OkYKA?t=3660) |\n| [marimo-pair](skills\u002Fmarimo-pair) | A coding agent drives a reactive Marimo notebook through a bash bridge into the Python kernel, for human-in-the-loop EDA. | Eric Ma (Moderna) | [00:11:57](https:\u002F\u002Fyoutube.com\u002Flive\u002Fl37PR-OkYKA?t=717) |\n\n| Workflow | What it does | Guest | Watch |\n|----------|--------------|-------|-------|\n| [agentic-eda](workflows\u002Fagentic-eda) | Human-in-the-loop EDA: agent renders the next plot, human picks the next question, every claim backed by an artifact. | Eric Ma (Moderna) | [00:23:27](https:\u002F\u002Fyoutube.com\u002Flive\u002Fl37PR-OkYKA?t=1407) |\n| [eval-driven-charts](workflows\u002Feval-driven-charts) | Build an agent-facing chart library by generalising eval failures into features; the package can never regress on an eval it once passed. | Bryan Bischof (Theory Ventures) | [01:25:11](https:\u002F\u002Fyoutube.com\u002Flive\u002Fl37PR-OkYKA?t=5111) |\n| [weekly-gremlins](workflows\u002Fweekly-gremlins) | Three agent personas pull from a bad-ideas backlog, pitch and critique each other, and write design docs for moonshots no roadmap would schedule. | Hilary Mason (Hidden Door) | [01:14:20](https:\u002F\u002Fyoutube.com\u002Flive\u002Fl37PR-OkYKA?t=4460) |\n\nStill to come from this episode:\n\n- **Local-first inference.** Tom Tunguz runs Qwen 35B on a Mac M5 at 120 to 140 tokens per second with a 256K context window; cloud only comes in for multi-file rearchitectures or hard bugs.\n\n## Episode 3: Matthew Honnibal, Eleanor Berger, Nico Gerold, Alan Nichol, Vincent Warmerdam, Paul Iusztin\n\n[Watch on YouTube](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs)\n\n| Skill | What it does | Guest | Watch |\n|-------|--------------|-------|-------|\n| [try-except](skills\u002Ftry-except) | Reads a Python codebase and tightens every `try\u002Fexcept` so the `try` covers only what can fail and the `except` catches the right exception. | Matthew Honnibal (spaCy, Explosion) | [00:12:09](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=729) |\n| [pre-mortem](skills\u002Fpre-mortem) | Reads production code, finds where it is fragile, and writes post-mortems for bugs that have not happened yet but a plausible change could introduce. | Matthew Honnibal (spaCy, Explosion) | [00:14:10](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=850) |\n| [mutation-testing](skills\u002Fmutation-testing) | Measures test-suite strength by introducing deliberate bugs one at a time and reporting which ones no test caught. | Matthew Honnibal (spaCy, Explosion) | [00:14:10](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=850) |\n| [here-now](skills\u002Fhere-now) | Publishes HTML pages, files, and whole sites to live URLs without leaving the terminal. | Eleanor Berger (Jimini Health) | [00:45:55](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=2755) |\n| [anki-connect](skills\u002Fanki-connect) | Drives Anki through the AnkiConnect API, gating every note- or card-modifying operation behind explicit confirmation. | Eleanor Berger (Jimini Health) | [00:49:46](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=2986) |\n| [impeccable](skills\u002Fimpeccable) | Hands a coding agent a full frontend design language so it builds production-grade interfaces instead of generic ones. | Eleanor Berger (Jimini Health) | [00:50:02](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=3002) |\n| [youtube-watch-later-gist-summaries](skills\u002Fyoutube-watch-later-gist-summaries) | Reads your YouTube Watch Later playlist, summarises every video from its transcript, and publishes each summary as a secret gist. | Eleanor Berger (Jimini Health) | [00:52:57](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=3177) |\n| [thread-postmortem](skills\u002Fthread-postmortem) | Introspects a thread that went sideways, traces each misstep to the instruction behind it, and proposes edits biased toward deletion. | Nico Gerold (Sourcegraph, Amp) | [01:59:04](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=7144) |\n| [remotion-video](skills\u002Fremotion-video) | Encodes a builder's design judgment for programmatic video, so Claude turns a few minutes of recorded audio into a finished explainer. | Alan Nichol (Rasa) | [02:46:00](https:\u002F\u002Fyoutube.com\u002Flive\u002Fud2WzkKeDZs?t=9960) |\n\nStill to come from this episode:\n\n- **Paul Iusztin's writing loop.** Diff a hand-edit against the agent's draft, extract the signal, and fold it back into a markdown style profile the agent reads next time.\n- **Vincent Warmerdam** showed notebooks as a shared canvas for humans and agents; his Marimo Pair skill already shipped in Episode 2 ([marimo-pair](skills\u002Fmarimo-pair)).\n\n## Upcoming episodes\n\nRegister on Luma to join live, or get the recording after.\n\n### Episode 4: Hamel Husain, Chris Fonnesbeck, Doug Turnbull\n\n[Register on Luma](https:\u002F\u002Fluma.com\u002Fltpzpqgw)\n\nJoined by Hamel Husain (Parlance Labs), Chris Fonnesbeck (PyMC Labs, veteran analyst for the Mets, Brewers, and Yankees), and Doug Turnbull (led Search at Shopify and Reddit).\n\n## More from us\n\nVanishing Gradients is a podcast, workshop series, blog, and newsletter focused on what you can build with AI right now. Over 70 episodes with expert practitioners from Google DeepMind, Netflix, Stanford, and elsewhere. Hundreds of hours of free, hands-on workshops. All independent, all free. [Subscribe on Substack](https:\u002F\u002Fhugobowne.substack.com\u002F).\n","该项目是一个配套的GitHub仓库，用于支持直播系列“Show Us Your (Agent) Skills”，旨在探索顶尖开发者如何使用AI代理进行开发。核心功能包括一系列可浏览、分叉和自定义的工作流示例、工具参考及具体的技能实现，如代码解释、GitHub回复自动化等。项目采用JavaScript编写，并通过npx命令行工具简化了安装与更新过程。这些技能来源于实际直播中的演示，便于开发者学习并应用到自己的项目中。适用于希望了解前沿AI代理技术及其在实际开发中的应用案例的软件工程师或团队。","2026-06-11 04:03:38","CREATED_QUERY"]