Skills
One skill, added once, makes your agent work differently.
By Mackenzie Bowes, with MBSI's agents.

Find the one for your work

iconBuilder
For agents that write and change code from a brief: a command, a site, a page or a migration.
5 skills →

iconImages
For agents that make what a reader sees before they read: headers, cards, figures and diagrams.
4 skills →

iconPlan and decide
For agents that plan what happens next, hand out work, and ask a person for a decision.
8 skills →

iconTools and checks
For agents that look after shared tools and checks: the small scripts that do the work and the checks that prove it.
2 skills →

iconProduct
For agents that shape goals into briefs and make things judged by readers, visitors or buyers.
5 skills →
Add them to your agent

The skills work in Claude Code, the Claude apps, ChatGPT and Grok. In Claude Code, add the whole collection, then any skill by name:
/plugin marketplace add mbsi-labs/skills/plugin install <name>@skills
Each skill's page has the other ways in.
How they fit together

Our agents work in kits, one for each kind of work. A skill sits in the kits whose work needs it.
How we make them

Most start as one person's idea. signal-chain is Mackenzie's model of how a reader decodes what is in front of them. cognitive-prophylactics is his idea that a story shows a reader what lifts or lowers their standing before they meet the situation. but-therefore is Trey Parker and Matt Stone's rule for stories, with Mackenzie's way of using it on devlogs. be-ambitious comes from Felix Rieseberg's tutorial, lever and first-principles from the pstack skills.
A named agent writes it up from his own words. Rowan wrote cognitive-prophylactics from three lines Mackenzie said while planning a game, and be-ambitious from his request: “derive a new skill called /be-ambitious that follows the ideas in the text”. Heron adapted lever and first-principles from pstack. Each skill's credits file keeps Mackenzie's words with the date, and says which parts the agent added and that they are his to correct. The NOTICE file that ships with each skill names its outside sources and their licences.
Inside is a method, not a list of rules. but-therefore gives a test: write the story as one-line beats, read each joint as “but”, “therefore” or “and then”, and every “and then” gets one of three repairs (find the real cause, cut the beat, move it). cognitive-prophylactics asks for one line before drafting: the situation the reader might meet and the move that lifts or lowers their standing in it. be-ambitious is four steps in order, ending with writing the task again.
A check keeps our machine out of it. Eleven portability rules turn away any skill that still points at our paths, our links or our crew, and show the line and the fix.
Why we publish them
“I measure why AI agents fail at real work, then ship the fix as a skill.”Mackenzie Bowes, from his bio
Until now there was no clean, easy way to hand those fixes to anyone else. Each skill here is a folder you can install.
“Men become poor in isolation and rich in community.”Mackenzie Bowes

Skills: one skill, added once, makes your agent work differently. Five groups: Builder, Images, Plan and decide, Tools and checks, Product. Add them to your agent with /plugin marketplace add mbsi-labs/skills. By Mackenzie Bowes, with MBSI's agents.
All skills, by the job they do
Each skill is a short set of instructions an agent follows for one job. Every page says who it helps, shows the same request answered without it and with it, and says how to install it where you work.
Think it through and decide
When you have too many options and a wrong pick would cost you.
- Make the choice, then show the result, not the method: It compares the options for each person a choice affects, including doing nothing, then shows you a short result you can check.
- Fold a new requirement into a design as if it was always there: When a rule changes, it redesigns the type model or module from the start instead of adding another special case.
Write something people read
When you have the facts but the draft reads like a list of things that happened.
- Turn a list of events into a story that people finish: Rebuilds a devlog or post as a chain of "but" and "therefore" instead of "and then".
- Write a story your reader can use: Before you draft, it asks for one line: the situation your reader might meet and the move that lifts or lowers their standing in it.
- Pick the format your reader can afford to take in: It ranks ways of carrying a message by how costly each is for the reader to decode, from faces and voices to writing and symbols, so you choose the one they can afford.
- Know why a piece exists before you write it: Before you write, it asks four questions in writing: the point of the piece, why a reader would care, the business goal, and what making it says about us.
Run a team of agents
When your agents keep forgetting your rules or repeating the same mistake.
- Get more from your agent than the first answer: Before your agent starts, it asks what the job is really for and what an expert would do, then asks for more than you first typed.
- Find the false belief behind a failure that keeps repeating: After a failure, it has your agent write what it did and what it believed, as axioms, principles and cruxes, then test each belief to find the one that is wrong.
- Fix a repeated mistake in structure, not with one more instruction: When the same correction comes up a second time, it picks the cheapest guard that holds it (a type, a lint rule, a hook or a script) instead of adding another line of text.
- Build the small tool that does the job, not the job by hand: When the work isn't trivial, it has your agent write the script that does it, rerun it on the first unit and diff the result against the hand version.
- Make your agent check the real thing before it says done: Before your agent says it is done, it runs the real program from outside and reads what it prints, instead of trusting a compile, a cached screenshot or its own word.
- Change your agent's behaviour with one small edit, not a long rulebook: A change to a skill, rule or hook ships only if it is small, one or two intentful edits, because every added line carries claims nobody checked.
- Keep your agent moving instead of asking about everything: Your agent proceeds on reversible work, settles the mechanics itself and shows what it did and why, and still asks first about anything hard to reverse, outward-facing or costly.
If you're working on something like this, I'd like to compare notes.
Agents, harnesses, internal AI tooling: send me what you're stuck on or what you've measured. I'll tell you what I've seen.
If you are working on something like this, Mackenzie would like to compare notes: mackenzie@mackenziebowes.com or @_mackenziebowes on X.