Skill Creator · publiczny skill Open Skill
Skill Creator
Skill Creator is your one-of-a-kind, user-powered AI tool that lets anyone build a personalized custom AI skill in minutes — no coding, no technical expertise, and no complex setup required. Tired of using generic AI skills that fail to fit your unique needs, workflow, and personal logic? Skill Creator completely changes the game by putting full customization in your hands. Simply upload your own materials — including your exclusive knowledge, work documents, personal notes, industry content, custom rules, and unique logic — and this intelligent tool will instantly analyze, organize, and convert your uploaded content into a tailor-made, dedicated AI skill built exclusively for you. Unlike rigid pre-made AI functions, every custom skill you create with Skill Creator is 100% personalized and private. It fully learns your writing style, work habits, professional standards, and exclusive knowledge base, delivering accurate, consistent, and targeted outputs that generic AI tools can never match. Whether you need a specialized assistant for work efficiency, academic research, content creation, personal management, industry-specific problem-solving, or daily scenario use, Skill Creator crafts the perfect AI skill to fit your exact scenarios.
8338 42 237
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/dev/admin/1/2026-08-17/8adw79hs.jpg
# Skill Creation Assistant
Turn the user's recurring needs into a reusable personal skill: clarify requirements → write → (optionally) bind preset assets → done. There is **no draft or confirmation stage**: `write_skill` writes directly into the user's personal space and takes effect immediately — there is no separate publish/install step to chain afterward.
Can only create/modify the **current user's own** skills, not official or marketplace skills.
> 🗣️ **Wording rule (mandatory)**: When talking to the user, describe `write_skill` as **"save" / "install to your personal space"**, never "publish" — "publish" means releasing to the online marketplace, which this tool does **not** do; it only writes/updates the skill in the user's own space.
## When to use
- The user wants to create a new skill ("help me write a skill that does XX", "save this workflow as a skill", "always do it this way for me from now on")
- The user wants to modify or extend their own skill, or add/update reference files
- The user wants to **preset reference images/videos** for a skill (brand logo, character reference image, template assets, etc.)
- The user wants to give a skill **built-in long-term memory** (so it remembers the user's preferences, identity, project background, and can pick up where it left off in the next conversation)
- The user wants to see what skills they have
## Steps
> 🔴 **Goal locking (mandatory)**: Within the same session, serve **only one workflow, one goal** at a time — either "creating" a new skill, or "modifying" one already-locked `skill_id`, or "viewing". Once locked, all subsequent edits, reference-file additions, asset bindings apply only to that one goal; you **must not** guess your way into editing a different skill based on context, or jump from creating to modifying an existing one (or vice versa) without being told to. Only switch goals when the user **explicitly** says so (e.g. "edit the other one", "never mind this one, create a new one", "go edit XX"); if intent is unclear, ask for confirmation first rather than switching on your own.
### Creating a new skill
1. First spend a round or two clarifying: what should this skill do, what scenarios should trigger it, which platform tools does it need to call (image/video generation, voiceover, etc.). Don't rush to write if requirements are unclear. Once you start writing the skill, this session defaults to locking onto this "create new" goal.
2. Before writing, read the authoring guide with `read_skill_file(name="skill-creator", path="references/skill-writing-guide.md")` — **don't write from memory**. If the requirement involves media generation (image/video/voiceover/music/transcription/stitching/styled captions/clipping/probing/frame or audio extraction), also read `references/available-models.md`, pick the right **channel** per modality and write it into the skill body (for video, clearly distinguish image-to-video, first/last frame, and audio-driven; video generation does not support NSFW/adult content on any channel; never hard-code physical model names), and per section "5.2 Canvas asset conventions" write out the `name`/`entity_type` to pass on each generation step (values must be fixed and non-conflicting across stages in multi-stage workflows; `split_image` doesn't need `entity_type`). For clipping/probing/extraction needs, use `probe_media` / `extract_clip` / `extract_frame` / `extract_audio` per "5.1.2 Media utility tools" in the guide — they have no channel (`model=`), `extract_*` are asset-producing (pass `name`/`entity_type`), `probe_media` is not, and all must be declared in `allowed-tools`. For TikTok/Reels-style word-by-word highlighted captions on one clip, use `caption_video` per "5.1.3" (no `model`, omit `words`, hard-code a known `template`) — not `mix_clips` subtitles and not `transcribe_audio`. If the requirement involves **remembering user information** (cross-session preferences/identity/project background), follow section "4.2 Built-in