Skill System

📖 Best for: All roles — people who want to package team SOPs, best practices, and lessons learned into reusable how-to manuals that AI can invoke across tasks

📖 Reading time: 4 minutes

📖 In one sentence: YingClaw’s reusable procedural knowledge module — wrap your team’s fixed processes, lessons learned, and best practices into a 4-part standard structure (trigger conditions + operation steps + common pitfalls + verification methods), callable from any session via auto-activation or explicit invocation. Skills complement code plugins: plugins extend AI capabilities (run shell, read/write files, call APIs); skills teach AI workflows (when to do, how to do, what to avoid, how to verify). Triggered by natural language ("do this with skill X") or explicit skill_create / skill_patch / skill_read tool calls. All roles, stored at workspace/skills/<skill-name>/SKILL.md, publishable to the Skill Marketplace for community sharing.

---

I. Core Value

Value	Description
Reusable	Write once, call many times — team SOPs no longer depend on word of mouth; package into skills for auto-execution
4-part standard structure	Trigger + steps + pitfalls + verification — all four required; reproducible and verifiable
Knowledge crystallization	Crystallize "lessons learned" and "best practice order" into team assets; new hires can use on day one
Encryptable & shareable	Sensitive skills stored encrypted; general skills published to Skill Marketplace for community sharing

II. Main Capabilities

1. Skill 4-Part Standard Structure

Part	Role	Example
Trigger conditions	When to use it	"Triggered when user asks to auto-publish to CSDN"
Operation steps	How to do it (with commands / code)	"Step 1: open browser; Step 2: ...; Step 3: submit"
Common pitfalls	What to watch out for + how to avoid	"Login captcha may expire; retry 3 times on failure"
Verification methods	How to confirm success	"Check whether the new article appears in the blog list"

All four required — without triggers it's useless, without steps it's not executable, without pitfalls failures are high, without verification you don't know if it worked.

2. Skill Creation

• Use the skill_create tool, name must be lowercase letters + digits + hyphens (e.g. bid-generator)
• Required: name / description / content / optional tags
• Auto-generates workspace/skills/<name>/SKILL.md + .agent_created marker
• Trigger conditions go in the description field (for auto-activation)

skill_create(
  name="csdn-auto-publish",
  description="CSDN auto-login + publish article — Trigger: user asks to auto-publish to CSDN blog",
  content="## Trigger Conditions\n...\n## Operation Steps\n...\n## Common Pitfalls\n...\n## Verification Methods\n...",
  tags=["auto-publish", "blog", "csdn"]
)

3. Skill Storage & Organization

• Path: workspace/skills/<skill-name>/SKILL.md (+ optional scripts/ resources)
• Management: skill_read view / skill_patch modify / skill_delete delete
• Encryptable: sensitive skills stored encrypted; skill_file_read auto-decrypts; cannot manually edit encrypted skills — must decrypt first

4. Skill Invocation (Auto + Explicit)

• Auto-activation: when user request matches the skill description, YingClaw auto-loads
  • e.g. "publish an article with the CSDN auto-publish skill" → auto-loads csdn-auto-publish
• Explicit call: say directly in conversation "use skill X to do Y"
  • e.g. "use intelliyou-docs-manager skill to push to master"
• Skill composition: complex tasks can activate multiple skills, executing serially per dependency

5. Skill Marketplace (Community Sharing)

• Skill Marketplace is YingClaw's community skill store
• General skills (CSDN auto-publish, Excel compare, CRM ops) contributed by the community
• Browse / search / install / rate, all in one
• Enterprise supports private marketplace (internal skills, not exposed externally)

6. Skill Update & Iteration

• skill_patch modifies existing skills: step optimization, new pitfalls, bug fixes
• Triggers the "stumble → update" cycle: new problems in real tasks get added to "Common Pitfalls"
• Supports append mode or full replacement; changes take effect immediately

7. Skills vs Plugins (Key Distinction)

Dimension	Code Plugin	Skill
Form	Code module (Python / Node)	Knowledge document (SKILL.md)
Role	Extends AI what it can do	Teaches AI how to do it
Examples	Shell tool, file I/O, API call	CSDN auto-publish, CRM ops SOP
Relationship	Complementary, not exclusive — skills can call plugins to complete steps; plugins provide low-level capability

Simple mnemonic: plugins are the hands (execute), skills are the brain (orchestrate).

8. Skill Encryption & Security

• Sensitive skills (company API / internal credentials / proprietary processes) stored encrypted with obfuscated filenames
• skill_file_read auto-decrypts; seamless in normal use
• Team sharing gives "call permission" only, not "modify permission" — prevents accidental changes

III. Typical Use Cases

Use Case 1: Content Ops Automation — "Publish an article with the CSDN auto-publish skill"

"use csdn-auto-publish skill to publish this markdown to my CSDN blog"
→ load skill → steps: open browser → login → rich-text editor → paste markdown → submit
→ verify: check whether the new article appears in the blog list

Repetitive content ops (multi-platform publishing, auto data scraping) wrapped in skills → 5-10× ops efficiency.

Use Case 2: Business System Ops — "Use the CRM requirement management skill to log a new requirement"

"use crm-requirement-mgmt skill to log this new requirement into CRM"
→ steps: open CRM → Requirement Mgmt → New → fill fields (title/desc/priority/owner) → save
→ pitfall: priority is a dropdown, must select "Medium" (don't leave default "Low")
→ verify: search by title in requirement list, confirm the new one appears

Internal business system operations (CRM / ticketing / finance / HR) crystallized as skills; new hires can execute on day one with zero training.

Use Case 3: Data Analysis — "Use the Excel multi-platform compare skill for competitor analysis"

"use excel-multi-platform-compare skill to compare phone prices on JD / Tmall / Pinduoduo"
→ steps: scrape 3 platforms → write to Excel → multi-sheet compare → generate chart
→ verify: Excel 3 sheets complete + charts render normally

Repetitive data tasks (competitor analysis, report generation, data cleaning) solidified as skills → one-click professional output.

Use Case 4: Team Knowledge Crystallization — "Turn our SOP into a skill"

"Our team's 'customer complaint handling process' is a standard SOP, write it as a skill so new hires can use it"
→ skill_create(name="complaint-handling-sop", content="4-part structure...")
→ all team members' sessions auto-available; new hires can do it on day one

Team SOP digitization — word-of-mouth gets lost; skills don't, and they're executable.

Use Case 5: Internal Process Packaging — "Use intelliyou-docs-manager to push to master"

"use intelliyou-docs-manager skill to push the modified docs to master"
→ steps: frontmatter self-check → git add → commit → push → check CI
→ pitfall: frontmatter fields cannot have backticks, would cause YAML parse failure
→ verify: GitLab pipeline status = success

Complex multi-step processes (push code, deploy service, generate report) solidified as skills; no more re-deriving steps every time.

IV. Usage Guide

Step 1: Use existing skills — say "use skill X to do Y" in conversation; YingClaw auto-loads and executes. Tasks whose description matches a skill's description also auto-activate.

Step 2: Create your own skills — use skill_create(name, description, content, tags): name is lowercase-hyphen; description clearly states "what + when"; content is the 4-part structure; tags aid marketplace search.

Step 3: Update a skill — use skill_patch: add new pitfalls to "Common Pitfalls" when you stumble; optimize "Operation Steps"; refine "Verification Methods".

Step 4: Share to the marketplace — publish general skills to Skill Marketplace: polish description + tags, add examples and screenshots, write version + changelog.

Step 5: Encrypt sensitive skills — set encrypted=True on creation for auto-encrypted storage.

V. Best Practices

• All 4 parts required — write to the 4-part structure; best readability and executability
• Trigger conditions must be explicit — description must clearly say "what + when" so YingClaw can auto-activate precisely
• Operation steps must be executable — each step with concrete commands / paths / tool calls; avoid vague "open browser then..." descriptions
• Common pitfalls must be real — write pitfalls you actually stumbled on, not hypothetical ones; that's what makes a skill valuable
• Verification must be quantifiable — write "check whether X appears as Y" — observable and testable verification; avoid subjective "see how it looks"
• One skill, one thing — keep skill granularity fine ("CSDN auto-login + publish" beats "all content ops") for easier composition and reuse
• Encrypt sensitive skills — anything involving company API, credentials, or proprietary processes must be encrypted; no plaintext
• Don't write skills as plugins — skills are "how to do" knowledge (process), not "what can be done" code (capability); if code solves it, don't wrap it as a skill