Henrik Söderlund (@henkis) header
Henrik Söderlund (@henkis) profile picture

Ex-founder | Ai Consultancy | 🛠️ Ad Tech & Analytics | 📣 Digital Advertising, Media Buying | 📊 Measurement & Data

Perth, Western Australia
Joined June 2007
Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)Photo by Henrik Söderlund (@henkis)
About 70% of the comments in our agency's codebase recorded decisions. Most are still beside their code. I had Claude Code agents check all 18,000 comment blocks using my /delete-comment skill. About 490 agents across two sessions. About 360 comments went into ADRs - 38 new records and 44 amendments. I used Matt Pocock's test. Is the decision surprising and hard to reverse? Or does it record an avenue we've already explored and rejected, so agents stop recommending it again? Most decisions don't need an ADR. A comment can hold a why the code cannot show. We also deleted about 2,000 comments whose decisions were already in docs. Keeping another copy beside the code gave us another place for it to go stale. I'm thinking about adding the skill to my Claude Code plugin pack - would anyone be keen on this? github.com/henkisdabro/wooks… Thanks to @mattpocockuk for posting about comments this morning.
3
3
44
What a banger @mattpocockuk - it's so helpful seeing your skills being used "in action", thank you for taking the time!
It's been nearly 6 months since I posted a 'watch me work' deep dive on my skills. Since then, they've become the most-starred skills repo in the world. So, here's 70 minutes of me building a 21,000-line PR, especially focusing on: - DDD with /grill-with-docs - Improving your codebase with /retro - How to do human review effectively Enjoy!
22
I've jumped on the test-pruning bandwagon. The spark came from @mazeincoding's /prune-tests (great name, so I kept it). I wanted my own take, steered towards @mattpocockuk's skills so it speaks the same language. The problem: coding agents write tests and rarely stop. Copies of other tests. Tests that can't fail. All green, all slowing down the next change. My version does one cheap 80/20 pass. Read every test, cut what guards nothing, and when in doubt, break the production code and see what goes red. Then I made it improve itself. 21 runs on our agency monorepo, TypeScript and Python. One agent prunes, a second tries to prove it deleted the only guard of something, and every gap they find feeds back into the skill's wording. Nothing went in without evidence from the runs. The result: - 1,958 tests pruned (17.5%) - wrong deletions per run down from 2.7 to 1.3, all 35 caught ones restored before merge - a dozen lines cut from the skill because no run ever needed them Try it, one line in Claude Code: /plugin install prune-tests --marketplace henkisdabro/wookstar-claude-plugins And grab "/message" while you're there. It drafts emails, WhatsApp, Slack and social posts with a live preview. I wrote this post in it. github.com/henkisdabro/wooks…
5
5
208
Henrik Söderlund retweeted
New skill: modern-web-guidance Your agent looks up Chrome's Modern Web Guidance, applies it to your code, then checks the result in a real browser Layout shift, reduced motion, dark mode, dialog focus, more Start here: npx skills add vercel-labs/agent-browser
40
105
16
1,825
139,742
Just shipped wookstar-claude-plugins 8.0.0: 31 plugins for Claude Code. The Google Tag Manager and GA4 skills are rebuilt as guardrails - a glossary, the pitfalls models hit, and what changed in 2026. Every other skill fact-checked against vendor docs. github.com/henkisdabro/wooks…
6
4
192
I've completely rewritten my message skill for Claude Code. You know the struggle. Your AI agent writes you an email or a message in the terminal, you copy it, paste it into Gmail or WhatsApp, and it lands wrong. Stray quote lines. Broken indentation. Hard line breaks in the middle of sentences. Markdown asterisks everywhere. So I fixed it. Now you just ask your agent to draft an email, a WhatsApp or a post, and a window called Paste Ready opens with your draft formatted for wherever it's going: - Gmail and Outlook, with spacing that looks like you typed it there - WhatsApp, Slack, Telegram and Discord, each in its own markup - X, LinkedIn, Bluesky, Reddit and more, with a live character counter - Plain text and clean Markdown for everything else 22 destinations in total. Pick one, press C, paste. That's it. A few extras I'm happy with: - Long X posts split into a numbered thread on a free account - Tables survive, even in WhatsApp on a small phone - Dollar columns line up properly in email - Works on macOS, Linux and Windows via WSL It's open source and free. You'll love it if you try it, and I'd like to hear what breaks: github.com/henkisdabro/wooks…
8
1
12
617
Run /doctor prompt-audit on your Claude Code setup. Today, ideally. I tidy my CLAUDE.md, skills and rules all the time. One of my repos has about 75 instruction files driving real production work, and I treat them like code. I assumed the audit would turn up a few old "think step by step" lines. There were hardly any of those. What it did find was instructions that had gone out of date while every test stayed green: - a safety limit written down at double the figure in the code - an onboarding step giving access to two people when the rule says three, so someone would have been missed every time - an example command that fails the first time you run it - a skill telling me to send a parameter current Claude models reject with a 400 - a safety rule pointing at a review step that doesn't exist (that one's a ticket now) - an override justified by "my collaborator doesn't commit". He has hundreds of commits. 30 high-confidence findings and 56 medium. A bit humbling, given I thought I was on top of this. My clean-ups were always about trimming. Nobody was checking the text was still true, and Claude follows it to the letter either way. I had a second model check the fix plan before shipping, and it caught two changes that would have broken things. If it tells you the skill is off, check skillOverrides. Mine was switched off by my own settings.
reminder to run `/doctor prompt-audit` in your Claude Code repos if you haven't already :)
4
111
My "ticket burndown factory" is playing along really well with the new @herdrdev herdr-core and herdr ui. Sandcastle-kit (based off of @mattpocockuk's 'sandcastle') now uses herdr-core's agent orchestration features and native skill, and integrates beautifully with the host claude code session via a claude mod. Check out the screenshots for a better view, and feel free to test it in your own setup and tell me what you think. henkisdabro.github.io/sandca…
87
My git pre-push hook was dumping ~210 lines of pytest progress dots into Claude's context on every push. Nothing broke, so it ran like that for months. I checked 3 months of my session logs: about 600k tokens of noise - progress dots, ✓ lines and colour codes. Claude re-sends its context every turn, so the dots alone added ~17M cached tokens. The fix: print one summary line when tests pass, and the full error when they fail. A passing push went from ~5k tokens to one line. Check your own setup. Paste this into Claude Code: "Look through my Claude Code logs (~/.claude/projects) for noisy command output: progress dots, test pass lines, colour codes, install logs. Show me the biggest sources and where they come from. Then make each print one line on success and the full error on failure."
6
6
304
Cracker of an idea, @mattpocockuk. So simple, so effective! Set mine up today. My nuance: a repo's hooks, MCP servers and guards only load when the agent starts inside it. So home is the conductor, dispatching a named-model worker into each repo, and daily domain work stays in its own folder.
A lot of folks ask me "how do you work with agents in multiple repos at once?" I have a very simple solution - the Home Repo I have a single repo (the home) where all of my skills, steering files, wiki etc live. I run my agent in the home repo, and then inside AGENTS.md I have pointers to the other repos: ``` For skills, clone mattpocock/skills to ~/repos/skills. For Sandcastle, clone mattpocock/sandcastle to ~/repos/sandcastle ``` This is extremely simple and incredibly effective. The home repo lets you: 1. Put all your steering files and tooling in one place 2. Have a single spot to run all your agents in 3. Make sweeping changes over multiple repos Talked about this in the course yesterday and folks are loving it (see below). I'm sure you will too. Enjoy!
33
Henrik Söderlund retweeted
run this on your repo trust me.
With the recent wave of vulnerability disclosures, it’s a good time to take a closer look at your own code. Turn your coding agent into a team of security auditors with the open-source skill that seeded Cloudflare’s vulnerability discovery harness. The 6 phase approach: 1. Map the codebase 2. Hunt for vulnerabilities 3. Challenge each candidate 4. Record the findings 5. Verify the findings 6. Report what’s confirmed, rejected, or still needs validation Start your hunt here: github.com/cloudflare/securi…
33
87
1
3,863
635,939
sandcastle-kit v0.20.0 🏰 Agent runs now live in the new Herdr app: ▸ status view opens beside you ▸ run state and a needs-you alert in the sidebar ▸ a terminal per sandbox, nested under the run Herdr TUI works as before. github.com/henkisdabro/sandc… henkisdabro.github.io/sandca…
1
2
72
Henrik Söderlund retweeted
Don't gaslight us. "I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry." If you didn't want to be misleading, you would have left out the '50% savings' claim on the $200 plan—at 10x the price of the $20 plan ($20 x 10 = $200), a 50% discount clearly implied we were getting 20x the usage.
I lead product for Claude subscriptions. Today we're renaming our Max plans, and I want to explain why. When we launched our top plan in April 2025, it had a single limit: 5 hours. It gave you 20x Pro's limit, so we called it Max 20x. When we added weekly limits to Max 20x, we set them to 10x Pro's (the 5-hour limit stayed at 20x). I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry. The Ultra plan keeps the weekly limit (10x Pro), but we're removing the 5-hour limit so you can keep building without getting blocked mid-session. This starts today and rolls out in stages. To help prevent fraud and abuse, a temporary limit applies until we've verified your account. You'll get an email when your limit is removed. We've got more cooking and can't wait to share it. Thanks for building with Claude 🙏
52
36
9
584
32,670
Henrik Söderlund retweeted
Why technical people are poor communicators
38
204
19
2,256
98,744
Henrik Söderlund retweeted
Today we announce 1Password for Agents platform, building with companies @Muse, SpaceX Super Intelligence Grok @Bot, @AnthropicAI’s @Claudeai, @OpenAI, @TownAI, @Browserbase, @Browser_use, @heydana_ai, Instinct, @usekernel, @Perplexity_ai, @tryramp, @Tomo, and @wajo_ai, with more integrations on the way. 🎉 We've heard from so many of our customers who want to use these new personal AI agents to help them get more done, but want more control over what the agent can access with the same trust and ease of use they get when using 1Password directly. 1Password for Agents gives your agents access without giving up your passwords. When an agent reaches a sign-in form, it requests a credential from your vault. If you approve the request, 1Password directly auto-fills the credential in the destination site so the agent can’t see or take possession of the credentials. The agent gets the task done, but it doesn’t get your passwords.. You stay in control. 1Password for Agents free and optional for personal and family accounts. Unless you opt-in to an agent connector, agents cannot touch your vault. ➡️Learn more: bit.ly/487da6J
646
1,094
902
10,576
1,864,499
Henrik Söderlund retweeted
Herdr UI just got a lot bigger 🐏 - Signed Windows installers + in-app updates. - A new activity bar. - Cross-project PRs and issues. - Azure devices. - Deeper WSL support. Here’s the release in two minutes. full article: nitter.cf/fabienpenso/status/211…
9
3
80
6,461
Henrik Söderlund retweeted
s tier advice
How to Use Writing to Sharpen Your Thinking:
2
56
1
1,084
74,562
Last night I realised $200 of Claude API credit was expiring in about 8 hours. By morning I had 72 issues filed across 4 of my repos, every one checked by a second agent. It cost $178, none of it my own money. Max 5x gets $100 a month, Max 20x $200, Team up to $500. Here's the prompt to find what yours could do: gist.github.com/henkisdabro/…
3
126