
Critic
- Updated June 17, 2026
- jdpedrie/critic
critic is a Claude Code skill in the AI & Agent Building category. Multi-agent fiction review system with Claude, Codex, and Gemini reviewers
Key points
- critic
- AI & Agent Building
- AI-coding skill
Critic by the numbers
- Data as of Jul 7, 2026 (Skillselion catalog sync)
/plugin marketplace add jdpedrie/critic/plugin install critic@criticAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Last updated | June 17, 2026 |
|---|---|
| Repository | jdpedrie/critic ↗ |
What it does
Multi-agent fiction review system with Claude, Codex, and Gemini reviewers
README.md
critic
A Claude Code plugin for multi-reviewer fiction critique. The plugin runs parallel reviews from Claude (as a subagent inside cowork), Codex (via the OpenAI Codex CLI), and Pi (via the Pi harness). It cross-reviews, synthesises into ranked issues with stable IDs, and saves to your vault for future reviews to build on.
The vault has to be an obsidian-storyline project. Scenes live in Scenes/, characters and locations in Codex/, worldbuilding in Research/. The critic reads scenes in act → chapter → sequence order, exactly as storyline's own "Export project" command does, and uses the rich frontmatter to scope reviews and pre-filter context.
Quickstart
# Build the server
cd critic/server
go build -o ../bin/critic .
# Load the plugin
claude --plugin-dir /path/to/critic
# Point critic at your storyline project (the folder containing Scenes/, Codex/, Research/)
/critic:settings vault_path /path/to/vault/MyNovel
# Configure Pi (optional)
/critic:settings pi_provider google
/critic:settings pi_model gemini-2.5-pro
The first review:
/critic:manuscript
Interactive. Pick which reviewers run, which Pi model, which steps. Then it runs straight through.
Documentation
- Goals. What the critic is for and what it deliberately isn't.
- Architecture. How the pieces fit together.
- Vault layout. What files live where and what critic reads or writes.
- Skills. Every
/critic:*command in detail. - Server. The Go MCP server: tools, vault layer, agent wrappers.
- Prompts. Every prompt template, who uses it, override rules.
- Workflows. Example sessions from a fresh project through a revision cycle.
Skills, briefly
| Skill | What it does |
|---|---|
/critic:manuscript [author note] |
Full manuscript review with interactive setup. Three reviewers, optional adversary, cross-review matrix, synthesis. |
/critic:review chapter <N> / scene <file> |
Four-role review (analytical, immersive, structural, adversarial) scoped to one slice. |
/critic:close-read scene <file> / chapter <N> / all |
Line-editor and copy-editor pass. Subagent per scene. |
/critic:downstream chapter <N> / scene <file> |
What breaks in later scenes after the edit. |
/critic:extract chapter <N> / scene <file> / entity <name> |
Reconcile prose against the Codex; maintains .claude/codex-inventory.md. |
/critic:summarize |
Per-chapter summaries to summary/. |
/critic:consult <question> |
Quick second opinions from Codex and Pi. |
/critic:rebuttal <issue-id> |
Rebut, defer, or accept a flagged issue. |
/critic:assess <issue-id> |
Deep-dive on a single issue. |
/critic:settings |
View and update plugin settings. |
/critic:help |
Skill reference. |
See docs/skills.md for the full reference.
Prerequisites
- Claude Code 1.0.33+
- Go 1.21+ to build the server
- obsidian-storyline plugin set up on an active project
- Codex CLI authenticated (
codex login). Optional, disable via/critic:settings codex_enabled false. - Pi CLI installed and authenticated. Optional, disable via
/critic:settings pi_enabled false.
Claude itself runs as a Task subagent inside cowork. No external process, nothing to configure.