Menu
Use when creating specialized subagents for opencode or the task tool - covers description writing for auto-delegation, tool selection, prompt structure, and testing agents
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Use when analyzing a large corpus of text, code, or data that exceeds a single agent's effective context - orchestrates parallel Worker subagents, Critic review subagents, and a final Summarizer subagent with task tracking and failure recovery
Use to decide what kind of basic agent you should use
Use when creating a new opencode plugin or setting up plugin structure - provides complete file organization, manifest format, and component definitions for commands, agents, skills, hooks, and MCP servers
Use when creating, releasing, or maintaining an opencode Plugin Marketplace - covers marketplace.json schema, version management, release checklists, changelog conventions, and validation to prevent sync drift between plugin.json and marketplace.json
Use when completing development phases or branches to identify and update AGENTS.md files that may have become stale - analyzes what changed, determines affected contracts and documentation, and coordinates updates
Use when creating or editing skills, before deployment, to verify they work under pressure and resist rationalization - applies RED-GREEN-REFACTOR cycle to process documentation by running baseline without skill, writing to address failures, iterating to close loopholes
| name | creating-an-agent |
| description | Use when creating specialized subagents for opencode or the task tool - covers description writing for auto-delegation, tool selection, prompt structure, and testing agents |
REQUIRED BACKGROUND: Read the writing-opencode-directives skill for foundational guidance on token efficiency, compliance techniques, and directive structure. This skill focuses on agent-specific patterns.
An agent is a specialized assistant instance with:
Agents are defined as Markdown files in an agents directory (~/.config/opencode/agents/ for the user) and dispatched via the task tool. The filename (minus .md) is the agent's name.
Create when:
Don't create for:
agents/
my-agent.md
Template:
---
description: Use when [specific triggers] - [what agent does]
mode: subagent
model: anthropic/claude-sonnet-4-6
---
# Agent Name
[Agent system prompt - who they are, what they do]
## Responsibilities
- Task 1
- Task 2
## Workflow
1. Step 1
2. Step 2
There is no name field — the filename is the identifier. mode: subagent makes the agent dispatchable via the task tool (rather than a primary agent you switch to).
The description field determines when the main agent auto-delegates to your agent. It's searched when matching tasks to available agents.
Format: "Use when [specific triggers/symptoms] - [what the agent does]"
Write in third person. Injected into system prompt.
# Bad: vague, no triggers
description: Helps with code
# Bad: first person
description: I review code for security issues
# Good: specific triggers + action
description: Use when reviewing code for security vulnerabilities, analyzing authentication flows, or checking for common security anti-patterns like SQL injection, XSS, or insecure dependencies
Include:
Length: Be specific, not verbose. A few sentences at most.
Agents get the standard opencode toolset by default:
| Tool | What the agent uses it for |
|---|---|
read | Reading files, analyzing code |
grep | Searching code patterns |
glob | Finding files by pattern |
edit | Modifying existing files |
write | Creating new files |
bash | Running commands, git, tests |
todowrite | Tracking multi-step workflows |
task | Spawning sub-agents |
webfetch / websearch | Research tasks |
Principle: Restrict what the agent doesn't need. Fewer capabilities = more focused behavior. Express restrictions as a permission: block in frontmatter (the deprecated tools: list should not be used):
# Read-only agent (e.g. a reviewer)
permission:
edit: deny
# Agent that must ask before running shell commands
permission:
bash: ask
Example restrictions:
edit: deny (no write access)edit: deny, relies on read/webfetch/websearchStart with who the agent is:
You are a security expert specializing in web application security and secure coding practices.
Explicit, numbered list:
## Your Responsibilities
1. Identify security vulnerabilities
2. Review authentication logic
3. Check for insecure dependencies
4. Report findings with severity ratings
Step-by-step process:
## Workflow
1. **Scan:** Use `grep` to find common vulnerability patterns
2. **Analyze:** Use `read` to examine flagged files
3. **Verify:** Use `bash` to run security audit tools
4. **Report:** Provide structured findings
Define expected structure:
## Reporting Format
For each finding:
- **Severity:** Critical/High/Medium/Low
- **Location:** `file:line`
- **Issue:** What's vulnerable
- **Impact:** What attacker could do
- **Fix:** How to remediate
What the agent should NOT do:
## Constraints
- Report findings only; do not modify code
- Ask for clarification if scope is unclear
- Escalate to human for ambiguous security decisions
| Model | Use For |
|---|---|
anthropic/claude-haiku-4-5 | Simple tasks, fast iteration, high volume |
anthropic/claude-sonnet-4-6 | Balanced capability/cost, most tasks |
anthropic/claude-opus-4-8 | Complex reasoning, critical decisions, code review |
Specify in frontmatter:
model: anthropic/claude-opus-4-8
Run the task WITHOUT the agent. Document:
Run with agent. Verify:
Test with:
If agent fails:
---
description: Use when reviewing code changes, pull requests, or verifying implementation quality - analyzes for bugs, style issues, and best practices
mode: subagent
model: anthropic/claude-opus-4-8
permission:
edit: deny
---
# Code Reviewer
You are a senior engineer reviewing code for correctness, readability, and maintainability.
## Responsibilities
1. Identify bugs and edge cases
2. Check error handling
3. Verify naming and style consistency
4. Suggest improvements
## Workflow
1. Read the changed files
2. Analyze for issues
3. Provide structured feedback
## Output Format
For each issue:
- **File:Line:** location
- **Severity:** Critical/Major/Minor
- **Issue:** description
- **Suggestion:** how to fix
---
description: Use when gathering information from the web, investigating APIs, or synthesizing documentation from multiple sources
mode: subagent
model: anthropic/claude-sonnet-4-6
permission:
edit: deny
---
# Research Agent
You are a research specialist gathering and synthesizing information.
## Responsibilities
1. Search for relevant sources
2. Extract key information
3. Synthesize findings
4. Cite sources
## Workflow
1. `websearch` for relevant sources
2. `webfetch` promising results
3. Extract and organize findings
4. Return structured synthesis with citations
---
description: Use when implementing specific tasks from plans - writes code, runs tests, commits changes following TDD workflow
mode: subagent
model: anthropic/claude-sonnet-4-6
---
# Task Implementor
You implement tasks following TDD principles.
## Responsibilities
1. Write failing test first
2. Implement minimal code to pass
3. Refactor if needed
4. Commit with descriptive message
## Constraints
- Never write implementation before test
- Run tests after each change
- Commit atomic, working changes only
| Mistake | Fix |
|---|---|
| Vague description | Include specific triggers and symptoms |
| Unrestricted permissions | Add permission: block (e.g. edit: deny) when the agent shouldn't modify files |
| No workflow | Add step-by-step process |
| No output format | Define expected structure |
| First-person description | Write in third person |
| Overly broad scope | Narrow to specific responsibility |
| No testing | Test auto-delegation and output quality |
mode: subagent set