A9 Building a Creative Workstation

💡 TL;DR: Create custom creative Agents, Skills, and shortcut commands to solidify your content production workflow.

📝 Course Notes

Key takeaways from this lesson:

---

What You'll Be Able to Do

• Create custom creative Agents (WeChat Official Account / Xiaohongshu / Novel / Screenwriter)
• Create creative Skills (topic planning / batch generation / style checking)
• Create shortcut commands (/wechat, /xiaohongshu, /polish)
• Build a personal creative workstation

---

Why Use OpenCode Instead of Web-based AI?

Capability	Web-based AI	OpenCode
Custom Agents	❌ Only generic AI	✅ Create custom personas, permanently effective
Shortcut Commands	❌ Re-enter every time	✅ /wechat triggers complete workflow instantly
Skill Reuse	❌	✅ Complex instructions packaged as Skills, call anytime
Workflow Persistence	❌	✅ Processes, styles, templates all persist
Team Sharing	❌	✅ Config files version-controlled, team reusable

This is the ultimate value of OpenCode — not just using AI, but building your own AI workstation.

---

Your Current Struggle

• Entering the same prompts repeatedly for every writing task
• Can't solidify common writing workflows
• Want a "dedicated writing assistant" that understands your style

---

When to Use This Technique

• When you need to: Solidify writing workflows and improve efficiency
• And don't want to: Teach AI your preferences from scratch every time

---

🎒 Before You Start

Make sure you've completed the following:

• [ ] Completed A1 Creative Workflow
• [ ] Completed 3.2 Understanding Agents
• [ ] Have a writing project folder

---

Core Concepts

Three Elements of a Writing Workstation

Element	Purpose	Example
Agent	Custom AI persona	Novelist Agent, Screenwriter Agent
Skill	Complex instruction packaging	Character design skill, Plot generation skill
Command	Quick trigger	/character, /scene, /polish

---

Common Agent Fields

Fields configurable in Frontmatter:

Field	Type	Description	Example
name	string (required)	Agent name, filename is the name
description	string	Agent function description (shown in menu)
mode	"primary" or "subagent"	Primary agent switchable via Tab, subagent called via @
temperature	number (0.0-1.0)	Controls randomness: 0.0-0.2 focused, 0.3-0.5 balanced, 0.6-1.0 creative
model	string	Override default model, format: provider/model-id
permission	object	Fine-grained permission control, e.g. { "edit": "ask" }
hidden	boolean	Hide subagent (not shown in @ autocomplete)
color	hex color	Custom Agent display color, e.g. #FF5733
steps	positive integer	Maximum iterations, prevents infinite loops
disable	boolean	Set to true to disable this agent
options	object	Extra model options passed to provider, e.g. OpenAI's reasoningEffort
disable	boolean	Disable this Agent

💡 All unlisted fields are passed directly to the provider as model options (e.g., OpenAI's reasoningEffort)

---

Follow Along

Step 1: Create a Writing Agent

Why
A dedicated Agent understands your writing style and preferences.

mkdir -p ~/.config/opencode/agent

Create an Agent config file. Note:

• Filename is the Agent name, e.g. novelist.md
• Use the mode field in frontmatter (not agent)

opencode

Help me create a writing Agent, save to ~/.config/opencode/agent/novelist.md with the following content:

---
name: novelist
description: Professional novel writing assistant, skilled in plot conception, character design, and text polishing
mode: subagent
temperature: 0.8
---

# Novelist Agent

You are an experienced novel writing mentor, skilled in:
- Story structure design (three-act structure, hero's journey)
- Character development (personality, motivation, arc)
- Plot progression (conflict, twists, climax)
- Text polishing (pacing, atmosphere, dialogue)

## Working Principles
1. Understand the author's intent before providing suggestions
2. Give specific, actionable suggestions with examples
3. Respect the author's style, don't impose your own views
4. Encourage innovation, avoid clichés

Step 2: Create a Writing Skill

Why
Skills package complex writing instructions for one-click invocation.

Important: Skill directory structure must be skill/<name>/SKILL.md

mkdir -p .opencode/skill/character-design

Create a character design Skill:

Help me create a character design Skill, save to .opencode/skill/character-design/SKILL.md:

> ⚠️ **Naming Rules (Recommended)**
> - Use lowercase letters, numbers, hyphens
> - 1-64 characters
> - Don't start or end with a hyphen
> - Don't contain consecutive hyphens (`--`)
> - Must match the directory name
> 
> Recommended format: `^[a-z0-9]+(-[a-z0-9]+)*$`

---
name: character-design
description: Generate detailed novel character cards
---

# Character Design Skill

Generate complete character cards based on user-provided character types.

## Output Format

### Basic Information
- Name, age, occupation, appearance

### Personality Spectrum
- 3 positive traits
- 2 negative traits

### Background Story
- Key events that shaped the personality

### Core Motivation
- External goal
- Internal need

### Character Arc
- Starting state → Ending state

### Signature Traits
- Catchphrases or mannerisms

Skill Permission Control

You can control which Skills can be used by agents through configuration. In opencode.json:

{
  "permission": {
    "skill": {
      "character-design": "allow",
      "internal-*": "deny",
      "*": "allow"
    }
  }
}

Permission	Behavior
allow	Skill loads immediately
deny	Skill hidden from agent, access denied
ask	Request user approval before loading

Wildcard patterns supported: internal-* matches all skills starting with internal-.

Step 3: Create Shortcut Commands

Why
Shortcut commands trigger common operations with a single keystroke.

mkdir -p .opencode/command

Create /character command:

Help me create a shortcut command, save to .opencode/command/character.md:

---
description: Quickly generate character cards
agent: novelist
---

Please use the character-design skill to design a character for me:

Character type: $ARGUMENTS

Create /polish command:

Help me create a shortcut command, save to .opencode/command/polish.md:

---
description: Polish selected text
---

Please polish the following content to improve expressiveness while preserving the original meaning:

$ARGUMENTS

---

Advanced Command Features

Besides $ARGUMENTS, the following advanced features are supported:

Positional Parameters

Access individual parameters:

Parameter	Description
$1	First parameter
$2	Second parameter
$3	Third parameter
...	And so on

Example:

---
description: Create file and write content
---

Create a file named $1 in directory $2 with content:
$3

Run: /create-file config.json src "{ \"key\": \"value\" }"

Subtask Mode

Use subtask: true to force command execution as a subtask, without polluting the main context.

---
description: Deep code analysis
subtask: true
---

Execute deep code analysis:
$ARGUMENTS

Shell Output Injection

Use !command to inject bash command output into command templates.

Example: Analyze test results

---
description: Analyze test coverage
---

Current test results:
!`npm test`

Suggest optimization based on the results.

Example: Review recent changes

---
description: Review recent changes
---

Recent Git commits:
!`git log --oneline -10`

Review and suggest improvements.

Step 4: Test the Writing Workstation

Why
Confirm configuration is working.

Restart OpenCode:

opencode

Test Agent:

Press Tab to switch, check if "novelist" Agent appears

Test command:

/character domineering CEO

You should see: AI generates a character card using the Novelist Agent and Character Design Skill

---

Advanced Configuration Tips

Permission Control

Agents can finely control tool and skill access permissions. Configure in the agent's frontmatter:

---
name: readonly-reviewer
description: Read-only code review
permission:
  edit: deny
  bash:
    "*": ask
    "git diff": allow
    "git log*": allow
  webfetch: deny
---

Only analyze code, no modifications.

Permission options:

• "ask" — Request approval before execution
• "allow" — Allow all operations without approval
• "deny" — Disable tool

Nested Agents

You can organize agents in subfolders, for example:

.opencode/agent/
  ├── writing/
  │   ├── novelist.md
  │   └── copywriter.md
  └── coding/
      ├── frontend.md
      └── backend.md

Then call them via @writing/novelist or @coding/frontend.

---

Checklist ✅

All items must pass before continuing

• [ ] Created a writing Agent
• [ ] Created at least one Skill
• [ ] Created at least one shortcut command
• [ ] Commands work correctly

---

Common Pitfalls

Issue	Cause	Solution
Agent not working	Config file in wrong location	Confirm it's in ~/.config/opencode/agent/ or .opencode/agent/, filename is the agent name
Agent frontmatter error	Used wrong field	Use mode in frontmatter not agent, required fields are name and description
Skill not found	Wrong directory structure	Confirm directory is skill/<name>/SKILL.md (lowercase), not skill/<name>/SKILL.md
Skill naming invalid	Naming rule violation	Lowercase alphanumeric, 1-64 characters, no leading/trailing hyphens, no consecutive hyphens
Command error	Parameter syntax error	Check usage of $ARGUMENTS, $1, and other placeholders

---

Lesson Summary

You learned:

1. Create custom writing Agents
2. Create Skills that package complex instructions
3. Create shortcut commands for one-click triggers
4. Combine all three to build a writing workstation

🎉 Congratulations on completing all writing track courses!

---

Next Steps

• Want more customization techniques? → Phase 5: Deep Customization
• Want to try other scenarios? → Developer Track or Productivity Track