Skill Creator
SkillAI & modelsGuide for creating effective skills. Use when users want to create or update a skill that extends agent's capabilities.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Skill Creator skill
What this skill tells your AI
The instructions your AI receives, as published by sikao-engine/kimix in kimi-cli/src/kimi_cli/skills/skill-creator/SKILL.md and read by ahel’s review.
Guidance for creating modular skill packages that extend Kimi's capabilities.
Core Principles
- Concise is Key — context is limited; only add info Kimi doesn't already know; prefer concise examples over verbose explanations.
- Degrees of Freedom — match specificity to task fragility: high freedom (text), medium (pseudocode/scripts with params), low (specific scripts needing consistency).
- Anatomy —
SKILL.md(required: YAML frontmattername+description, optionaltype; body under 500 lines) + optional resources:scripts/(executable code),references/(large docs, loaded on demand — keep >10k-word files here),assets/(templates/images). Do NOT include README.md/CHANGELOG.md.
Progressive Disclosure
- Metadata (name + description) — always in context
- SKILL.md body — when skill triggers
- Bundled resources — as needed
Keep SKILL.md lean; link reference files with clear guidance on when to read them; avoid deeply nested references.
Locations
Priority (most specific wins): --skills-dir → project (.kimi/.claude/.codex/.agents skills) → user (~/.config/agents/skills, ~/.agents/skills, ~/.kimi/skills, ...) → built-in. Brand-specific dirs (.kimi, .claude, .codex) beat generic (.agents, .config/agents).
Supported Forms
- Subdirectory (canonical):
<skills-root>/<skill-name>/SKILL.md— use for bundled resources. - Flat:
<skills-root>/<skill-name>.md— single-file skills; stem = name whennameomitted.
Creation Process
- Understand → 2. Plan (identify reusable resources) → 3. Initialize (dir + SKILL.md + resource folders) → 4. Edit → 5. Validate (frontmatter, naming, structure, discovery) → 6. Iterate.
Naming: lowercase letters/digits/hyphens, <64 chars, verb-led (e.g. gh-address-comments); folder name matches skill name.
Frontmatter: name, description (what + when to use); type: flow adds a mermaid/d2 block.
Body: imperative form; multi-step workflows with decision points; output formats/quality standards; links to reference files.
Testing & Validation
Test scripts (representative sample for many similar ones); verify discovery with kimi --skills-dir <parent-root>. Before complete: frontmatter starts/ends with ---; name matches dir/stem; description present (what + when); type is standard or flow (omit for standard); flow skills have a valid mermaid/d2 block; no README.md/CHANGELOG.md; resources referenced clearly from SKILL.md.
Signals
- GitHub stars
- 111
- Forks
- 12
- Last commit
- Sep 2026
- Hacker News mentions
- 1
Advanced
- Catalog kind
- skill
- Gateway key
skill-creator-sikao-engine- Source
- github.com/sikao-engine/kimix