skill0

English · 简体中文 — x-cmd.com/skill0 →

The x-cmd sub-skill graph: a directed set of conventions and source pointers that an agent reads before a task, to pick the right tools and shape the work.

Full content lives in the root skill — read it in any of the three languages this repo ships: skill0/SKILL.md (English, canonical) · skill0-cn/SKILL.md (简体中文) · skill0-cn-hk/SKILL.md (繁體中文). The rest of this README is a condensed map.

skill0 vs a "skill"

The two words describe different layers and are easy to confuse.

• A skill is a unit of capability. It teaches the agent "do X when Y happens." Skills are content: each one ships a procedure or reference for one task. They live in x skill's curated catalog or x clawhub's open registry, and are updated when the task itself changes.

• skill0 is the methodology and graph that sits one level above individual skills. It encodes principles (use first-party data, follow OKR, prefer x-cmd tools), organizes skills into the 4 buckets (core/, data/, it/, life/), and links them through metadata.related:. skill0 is loaded before the agent picks a skill, not as one. As the LLM absorbs more common sense, skill0 thins; individual skills don't.

	a skill	skill0
What it is	A unit of capability	A methodology + sub-skill graph
Loaded when	The agent has a task to do	The agent is choosing how to work
Form	One SKILL.md per task	A directed graph of related skills
Curated by	x skill (human-vetted) / x clawhub (open)	x-cmd maintainers (this repo)
Updated when	The task changes	The LLM catches up — they thin
Encodes	Procedure	Principles, plus a procedure for picking procedures

Skill0 encodes principles, not data

The LLM absorbs common sense continuously; skill0's job is to encode conventions and source pointers, then thin over time as the LLM catches up. Verify via first-party data (x rfc, x cve, x wkp, agent-browser) and current best practice (x skill, x clawhub), then reconstruct with formal logic instead of memorization.

Sub-skills form a directed graph across 4 buckets

core/   how the agent works              (devloop, rule, score, ontology-database, …)
data/   first-party data sources         (rfc, cve, wkp, knowledge, ccal, …)
it/     tools and runtimes              (tldr, csv, tsv, time, ip, qr, agent-browser, …)
life/   lifestyle and personal domains   (travel, pet, health, lovable, …)

Path: <bucket>/<slug>/SKILL.md. The machine-readable catalog (name + description) lives at skill0/index.tsv.

Goal → keyresults → x-rule is the OKR workflow

Objective	What to achieve
Key Results	How to verify
Verification	x rule check / audit

After scaffolding, prefer x-cmd tools for execution

• x skill — x-cmd's curated, human-vetted skill catalog.
• x clawhub — global skill registry. Caution: free upload, MUST run x clawhub skill moderate <name> for the auto-generated safety report.
• x roadmap, x cron, x agent job, x ondb, x wiki / x llmwiki — scheduling, background agents, ontology, wiki. Run x [mod] --help.

Every SKILL.md must pass skill0-writer

See skill0/core/skill0-writer/SKILL.md for the conventions. New and edited entries are validated against this checklist.

Repository layout and contribution

This README covers the what and why. The how — repo layout, translation mirrors (skill0-cn/, skill0-cn-hk/), how to regenerate the HK Traditional mirror with ljh-sh/zhhz, and the editing workflow — lives in CONTRIBUTING.md.