mark's Agent Skills

an attempt to write skills with high-signal activation because i didn't want to ask the agent to not use head on my windows machine, among other things

skills

core standards (always start here)

• dev-standards-majo — universal principles: British English, error handling, exit codes, AGENTS.md maintenance
• agents-md-authoring-majo — writing effective AGENTS.md files (see the 6-core-areas framework)
• skill-authoring-majo — creating new skills from patterns

language-specific

• python-majo — Python development with UV, basedpyright, ruff, Python 3.10+ syntax, and meadow Docstring Format (MDF)
• js-bun-majo — JavaScript/TypeScript using Bun (not npm, i've hopped on the Bun train)
• shell-majo — POSIX shell scripting (pure sh, not bash), error handling, exit codes
• csharp-unity-majo — C# development for Unity: naming conventions, British spellings, callback patterns

workflow and tooling

• git-majo — my git workflow: auto-commit after every prompt, conventional commits, never auto-push
• task-planning-majo — planning workflow for complex tasks: gather context, draft plan, 3 questions max
• writing-docs-majo — writing documentation in my voice: first person lowercase "i", casual but technical, British English
• mdf-majo - meadow Docstring Format (MDF) for Python
• mdf-md-api-docs-majo — Writing API refs or docs from code using the MDF

platform and setup

• running-windows-commands-majo — working on Windows: maps Unix commands to PowerShell equivalents (tail → Get-Content, etc.)
• setting-up-public-domain-repos-majo — setting up repos with dual licensing (Unlicense OR 0BSD)

sizes

trying not to overbloat the skills as per best practices

skill	lines	words	status
agents-md-authoring-majo	401	1748	🟢
csharp-unity-majo	210	914	🟢
dev-standards-majo	308	1374	🟢
git-majo	201	798	🟢
js-bun-majo	71	398	🟢
mdf-majo	444	1674	🟢
mdf-md-api-docs-majo	270	891	🟢
python-majo	335	1281	🟢
running-windows-commands-majo	443	1532	🟢
setting-up-public-domain-repos-majo	350	1613	🟢
shell-majo	450	1583	🟢
skill-authoring-majo	436	1852	🟢
task-planning-majo	265	1250	🟢
writing-docs-majo	490	1451	🟢

run count.py to update this table.

for whats in a skill: https://agentskills.io

how skills activate

skills load in three levels:

1. discovery — only name + description loaded at startup
2. activation — full SKILL.md loaded when the agent thinks it's relevant
3. deep dive — references/ and examples/ loaded when explicitly referenced

common triggers:

• "writing Python code" → python-majo
• "this is complex" → task-planning-majo
• "commit" → git-majo
• "README" → writing-docs-majo
• "Windows" → running-windows-commands-majo

using multiple skills together

skills stack. i usually do:

1. base: dev-standards-majo (always)
2. language: python-majo, js-bun-majo, etc.
3. workflow: git-majo, task-planning-majo (as needed)

design principles

• clear activation triggers — the description is the main signal. be specific: "Python with UV" not "a skill about code"
• actionable instructions — tell the agent HOW ("use uv add") not what ("uv is a package manager")
• explicit boundaries — always have "when to use" AND "do NOT use" sections
• test before shipping — check that the skill triggers when it should and doesn't when it shouldn't

skills vs AGENTS.md

these work alongside AGENTS.md (project-specific knowledge):

• use AGENTS.md for: project-specific stuff, exact flags, version-matched docs
• use skills for: cross-project patterns, large content, workflows

AGENTS.md wins because it's always loaded. skills win for stuff that doesn't fit.

validating skills

for my own use lol:

for /d %i in (majo-skills\*) do @echo Validating %i... && uvx --refresh --from ./agentskills/skills-ref skills-ref validate "%i"

licence

all skills are dual-licensed under The Unlicense OR BSD Zero Clause License (SPDX: Unlicense OR 0BSD).

go ham.