Skill v1.0.0
currentAutomated scan100/100name: ai-agent-skill-crafting description: Core AI agent meta-skill for adapting, sourcing, creating, structuring, and refining modular knowledge skills for Antigravity AGY, Gemini CLI, Hermes, and custom AI agents. metadata: version: "1.0.0"
🧠 Core AI Agent Skill Crafting & Refining Meta-Skill (ai-agent-skill-crafting)
This meta-skill defines standard operating procedures and core principles for AI agents to adapt, source, author, structure, and continuously refine specialized skill packages across workspaces.
🎯 1. The Skill Life-Cycle Pipeline
graph TDSource[1. Source Knowledge - Audit Repos & Codebases] --> Structure[2. Structure Spec - YAML Frontmatter & Markdown]Structure --> Author[3. Author Content - Commands, Diffs, DSOs & Rules]Author --> Verify[4. Verify & Adapt - Test Executables & Sub-Skills]Verify --> Distribute[5. Distribute & Sync - Local System & GitHub Workspace]
🔍 2. Sourcing Knowledge from the Core
Before writing a skill, agents must gather empirical evidence directly from authoritative source files:
- Codebase Inspection: Read exact method signatures, prototypes (e.g.,
piuu_core.h,LibC.kt), and build configurations (CMakeLists.txt,build.gradle). - CLI & Help Flag Execution: Test binary flags non-destructively (
agy-backup --help,gh run view --help). - Environment Audit: Inspect path variables (
$PREFIX/bin), shebangs, Python interpreters, and git remote URLs.
📝 3. Standardized Skill Specification & Metadata Structure
Every skill MUST be saved inside its own dedicated directory (user-skills/<skill-name>/SKILL.md) with clean YAML frontmatter:
---name: skill-identifier-namedescription: Concise 1-2 sentence description explaining the purpose, scope, and target integration of this skill.---# 🎨 Skill Title (`skill-identifier-name`)## 📌 1. Scope & Overview[High-level explanation of problem domain and solution]## ⚡ 2. Executable Code & Commands
Copy-pasteable, robust shell commands
## 🏗️ 3. Architecture & Design Principles[Diagrams, structural guidelines, edge-case prevention]## 🧪 4. Verification & Testing Strategy[Verification steps, assertion checks]
🔄 4. Adapting & Internalizing Skills
When an AI agent reads a SKILL.md specification:
- Extract Core Directives: Identify mandatory rules (e.g. "Always perform whole-environment backup" or "Do not auto-push without confirmation").
- Execute Sub-Skills: If the skill defines sub-skill verification routines or calculation foot-printing, execute the sub-skill pipeline completely.
- State Persistence: Update system memory files (
MEMORY.md,AGENTS.md) to retain knowledge across session resets.
🧹 5. Distribution & Refinement Workflow
To publish or update a skill across system environments:
# 1. Write skill specification in workspace repository# File: $HOME/Workspace/ai-agents-skills-workspace/user-skills/<skill-name>/SKILL.md# 2. Run the organization script to automatically sort skills into user-skills or system-skillscd $HOME/Workspace/ai-agents-skills-workspacepython3 organize_skills.py# 3. Sync to local active agent CLI skill directoriesmkdir -p ~/.gemini/antigravity-cli/skills/<skill-name> ~/.gemini/antigravity-cli/builtin/skills/<skill-name>cp -r $HOME/Workspace/ai-agents-skills-workspace/user-skills/<skill-name>/* ~/.gemini/antigravity-cli/skills/<skill-name>/cp -r $HOME/Workspace/ai-agents-skills-workspace/user-skills/<skill-name>/* ~/.gemini/antigravity-cli/builtin/skills/<skill-name>/# 4. Commit and push to remote skills repositorygit add .git commit -m "feat(skills): add/organize <skill-name> core skill"git push origin main
🔒 6. Global Privacy & Data Isolation Rule
When creating or refining skills, agents MUST adhere to strict data privacy standards:
- Zero Personal Data Exposure: NEVER hardcode personal user tokens, private account credentials, machine-specific account switches (
gh auth switch --user ...), or private SSH keys insideSKILL.mdspecifications. - Portable Environment Referencing: Use dynamic environment variables (
$HOME,~,$PREFIX) instead of hardcoded environment paths. - User-Centric Data Awareness: Direct the agent to interact with the user's data dynamically rather than embedding static personal identifiers.
🛠️ Required & Associated Agent Workspace Tools
When authoring, formatting, or discovering skills, activate these tools from `AI-Agents-Workspace-Tools-Library`:
- `wc-json-prompt` (`docs`): Extracts and repairs unformatted JSON payloads embedded inside natural language prompts.
- `wc-json-schema` (`docs`): Generates standard JSON schemas for structured skill inputs/outputs.
- `wc-tool-registry` (`docs`): Queries and indexes tools across all workspace categories.
- `wc-search` (`docs`): Performs noise-free symbol and keyword discovery across repositories.
- Upcoming Tool Note:
wc-skill-pack(AutomatedSKILL.mdYAML frontmatter linter and skill bundler) is scheduled for implementation.