The Publisher agent orchestrates the blog content pipeline. It chains four subagents through a linear pipeline: research, write, fact-check, review. Like Pai for the C-suite, the Publisher coordinates content agents and bridges context between stateless sessions.

Goal

Produce publication-ready blog posts by orchestrating the content pipeline end to end.

Tools

• Read / Glob / Grep / Write — direct file access
• Bash — invoke subagents

Subagents

Subagent	Status	Role	Model
Researcher	Active	Gather sourced facts, return research brief	Haiku
Writer	Active	Draft blog post from research brief	Sonnet
Fact Checker	Active	Verify claims against primary sources	Haiku
Reviewer	Active	Check style and structure against style guide	Haiku

Pipeline

graph LR
    R["Researcher"] --> W["Writer"]
    W --> FC["Fact Checker"]
    FC --> Rev["Reviewer"]
    Rev -->|needs revision| W

1. Researcher gathers sourced facts on a topic (read-only, web search)
2. Writer drafts a blog post from the research brief (reads style guide first)
3. Fact Checker verifies every claim against sources
4. Reviewer checks style and structure against the style guide
5. If issues are flagged, the Writer revises and the cycle repeats (up to 3 rounds)

Dual definitions

Content agents exist in both Claude Code and OpenCode:

Agent	Claude Code	OpenCode
Researcher	.claude/agents/researcher.md	.opencode/agents/blog/researcher.md
Writer	.claude/agents/writer.md	.opencode/agents/blog/writer.md
Fact Checker	.claude/agents/fact-checker.md	.opencode/agents/blog/fact-checker.md
Reviewer	.claude/agents/reviewer.md	.opencode/agents/blog/reviewer.md

Invocation

# Claude Code
claude --agent publisher
claude --agent researcher
claude --agent writer
claude --agent fact-checker
claude --agent reviewer

# Example prompts
# "Write a blog post about building MCP servers in Python"
# "Research the current state of AI code review tools"
# "Fact-check the draft at apps/blog/blog/markdown/posts/my-post.md"
# "Review the draft at apps/blog/blog/markdown/posts/my-post.md"