Markdown behaviors for the hako suite — and any agent that reads Agent Skills. Drop a folder, name it, the agent loads it.

_{site catalog  ·  SKILL.md spec  ·  hako-code  ·  hako-edit  ·  hako-studio  ·  org}

_{—— I ——}

Catalog

Pull only what you want — each is a self-contained folder, pure markdown, works in any agent or LLM.

— 01 —	corp A company in a folder. CEO · DEV · DESIGN · QA · ARCH — routes tasks by domain. Ships as CLAUDE.md + AGENTS.md, molds to its host.
— 02 —	brief Token-lean output. Cut ~50–75% of length, keep every fact. Drops filler, not information.
— 03 —	memory Durable cross-session memory: a MEMORY.md index + one-fact-per-file, with recall & dedupe rules. No database.
— 04 —	sync Change once, update everywhere — docs, version strings, sibling repos, JSON. Kills the "did I get them all?" miss.
— 05 —	scope Vague ask → goal + explicit non-goals + smallest viable step, before building. Stops scope creep.
— 06 —	release Bump version consistently, write the changelog, generate a multi-repo push-order checklist.
— 07 —	handoff Write a START-HERE state doc (done / next / gotchas) so the next session resumes in seconds.
— 08 —	changelog Turn a diff or session log into one clean CHANGELOG entry, in the project's own voice. Single-repo; the focused half of release.
— 09 —	ship The pre-push gate: build matrix · ASan/valgrind · version-drift · CHANGELOG sync · cruft · docs. Returns PASS or BLOCKED, never a vibe.
— ·· —	your skill here Open a PR. Folder name becomes the invocation.

_{—— II ——}

Install

From inside hako — the CLI, the editor's Rei pane, or hako studio:

:skill install mithraeums/skills          # all of them, into ~/.hako/skills
:skill install mithraeums/skills/brief    # just one
:skills                                   # what is loaded
:use brief                                # one skill as a standing mode, until :use none

Or by hand — each skill is a folder, so copying it is installing it:

git clone --depth 1 https://github.com/mithraeums/skills /tmp/skills
cp -r /tmp/skills/<skill> ~/.hako/skills/

hako looks in ./.hako/skills, ./.claude/skills, ~/.hako/skills and ~/.claude/skills, nearest first. Each skill costs one line of the prompt (its description) until the model needs it; the body and inner files are read on demand with the read_skill(skill, path) tool.

_{—— III ——}

Format

Every skill is a folder:

<skill-name>/
├── SKILL.md          dispatcher: frontmatter + intro + agent map
└── (any *.md, agents/, helpers, assets)

SKILL.md frontmatter:

---
name: <skill-name>
description: >
  One-paragraph summary. When to load. What it solves.
  Include trigger phrases.
---

Inside SKILL.md: keep it dispatcher-shaped — orient the model, list which inner files exist, hand off. Under 50 lines is the goal. If a skill needs more, split it into agents.

_{—— IV ——}

Compatibility

Tool	Path	Status
hako-code (hako)	~/.hako/skills/, .claude/skills/	native — descriptions up front, read_skill for the rest
hake (Rei pane) · hako studio	same as hako	they run hako, so they get the same skills
Claude Code	.claude/skills/	drop-in
Codex CLI · Cursor · Gemini CLI · OpenCode · Windsurf	varies	markdown is markdown

_{—— V ——}

Contribute

git clone https://github.com/mithraeums/skills
git checkout -b add/your-skill
# add your-skill/SKILL.md (+ any agents/helpers)
git push origin add/your-skill
# open PR

Keep skills referential, not prescriptive. Token budget first. The dispatcher orients — the model reasons.

_{— SEE LICENSE —  ·  GPL-3.0  ·  CONTRIBUTING}

_{— deus sol invictus mithras —}