Building My Own Agent Skills Catalog

The same speech, every session

Every fresh agent session used to start with the same speech: how I like documents structured, what my commit messages look like, which words I never want to see in a draft. The agent would follow along politely, and the next morning a new session would know none of it. My standards lived in chat history, and they evaporated with the scrollback.

Skipping the speech was worse. I once asked a bare agent for a research note and got back an eighteen-section format it invented on the spot. Another time I asked one to make a draft "more human," and it fabricated a detail that was never in the input: a claim that the logs sit in the Actions tab. Plausible, specific, and false.

So the choice was repeating myself forever or trusting output I had already caught lying. I wanted my rules written down once, versioned, and loaded automatically into every session. That's exactly what Agent Skills are for, so I published my own catalog:

github.com/su-ekachai/skills: my writing and analysis standards, packaged as Agent Skills anyone can install.

The skills I use in my workflow

A skill is a folder with a SKILL.md file in it: YAML frontmatter on top, instructions below. The agent reads the description, decides the skill fits the task, and loads the rest. The format is an open standard, simple enough that people publish whole catalogs of them, and I use other people's skills every day:

• superpowers brings process discipline: brainstorming before building, test-driven development, systematic debugging, and a strict "no completion claims without fresh verification" rule.
• The GSD suite runs my larger projects as phases: discuss, plan, execute, verify, with state that survives between sessions.
• ponytail pushes every solution toward the laziest version that works. It's the skill that argues against code.
• caveman does for replies what ponytail does for code: cuts them down to the point. I run it in lite mode, terse without going full grunt.
• taste-skill handles UI design. It's an anti-slop frontend skill that keeps my pages from coming out with the templated look every bare agent produces.
• ui-ux-pro-max, installed straight from the Claude Code plugin marketplace, backs my UI/UX choices with a reference library: 50+ styles, 161 color palettes, 57 font pairings, and 99 UX guidelines.
• mattpocock/skills carries my two favorites: grilling, which interviews you relentlessly about a plan until it stops being vague, and handoff, which compacts a session into a brief the next session can pick up.

The ones I wrote

The gap is that nobody ships a skill for how I write or how I read a company. Those standards are mine, so the skills had to be too. The catalog is MIT licensed and currently holds a writing category and a finance category:

• technical-writing: 12 rules for docs, commit messages, and PR descriptions, applied as write mode or as a review that reports rule-by-rule hits.
• blog-writing: keeps posts here sounding like me. Its voice profile was calibrated on my own published writing, and it checks drafts against a 47-item list of AI writing patterns before anything ships.
• fundamental-analysis: turns "analyze this ticker" into a structured research note: business model, financial health, moat, risks, and fair value from several models (DCF, PEG, Graham, DDM) with a scored verdict.
• crypto-news-screener: sweeps crypto news into a deduplicated report, each story scored 1 to 10 for impact with the affected coins and expected direction.

The writing skills guard how I sound. The finance skills encode how I evaluate. Every one of them started as a speech I was tired of giving.

One command, any machine, any agent

The part I underestimated is what publishing the catalog as a repo buys. Everything installs with one line:

npx skills@latest add su-ekachai/skills

On a new laptop, that command restores my entire standard in seconds. And the installer targets the shared ~/.agents/skills/ directory, then links each tool to it, so the same catalog serves every coding agent I run: Claude Code, Codex, opencode, and Antigravity. That works because Agent Skills is an open standard rather than one vendor's feature. I dropped one CLI for another this year and my skills didn't notice.

The repo cuts both ways, too. Anyone can install my catalog with that one command, the same way I installed superpowers and mattpocock's collection. My writing rules are now the same kind of artifact as the open-source skills I depend on: public, versioned, and one git log away from showing exactly when a rule changed.

Still growing

The catalog holds the skills I needed first; more will land as I keep noticing what I repeat. Every speech I catch myself giving an agent twice is the next skill waiting to be written.