Agents often drift and waste time on the wrong work. Correction Guy watches from the sidelines and steers them back on track.

Correction Guy keeps the agent honest. It reviews tool batches for drift and bad assumptions, and holds the agent back from stopping with unfinished work.
It talks to Codex directly through the Codex SDK, no other plugin required.
Works in Claude Code and Cursor.
Live monitor. Every few completed tool batches, a Codex pass scans the recent transcript for instruction violations, unsupported assumptions, and drift. Anything off is surfaced in the session and fed back before the agent continues.
Session prelude. At session start, the agent is reminded to check memory, restate the task, surface useful tools, verify third-party behavior before editing, finish the full work, and run the code.
Stop check. When the agent tries to stop, a Codex pass reviews the last response and returns a verdict. A minor issue is surfaced as an advisory message; a serious failure (asking permission instead of delivering, a stub, abandoned work, ignored review feedback, or skipping a required run) blocks the stop and feeds the correction back so work continues.
On-demand reminder. A /correctionguy command (Cursor) or /correctionguy:correctionguy slash command (Claude Code) reminds the agent to understand the task, do the full work, run the code, and ask for candid review.
By default the hook reviews run Codex in a read-only sandbox. Set CORRECTIONGUY_YOLO=1 to drop the OS sandbox: Codex threads get full filesystem and network access, with their limits conveyed by prompt text instead.
codex login once, or set an API key./plugin marketplace add correctionguy/correctionguy
/plugin install correctionguy@correctionguy
Install from the Cursor Marketplace when listed, or use local testing from this checkout:
mkdir -p ~/.cursor/plugins/local/correctionguy
rsync -a --delete --exclude .git --exclude node_modules ./ ~/.cursor/plugins/local/correctionguy/
cd ~/.cursor/plugins/local/correctionguy
bun install
Restart Cursor or run Developer: Reload Window, then confirm the plugin and hooks appear under Settings, Plugins and Settings, Hooks.
All knobs are environment variables; everything else is hardcoded.
| Variable | Default | Description |
|---|---|---|
CORRECTIONGUY_MODEL | gpt-5.5 | Codex model used for live-monitor and stop reviews |
CORRECTIONGUY_MODEL_REASONING_EFFORT | xhigh | Reasoning effort: minimal, low, medium, high, or xhigh |
CORRECTIONGUY_SERVICE_TIER | fast | Codex service tier (e.g. fast, flex, priority) |
CORRECTIONGUY_FAST_MODE | true | Enable Codex fast mode (true/false, 1/0, on/off) |
CORRECTIONGUY_MONITOR_EVERY_BATCHES | 3 | Completed tool batches between live-monitor reviews; 0 disables it |
CORRECTIONGUY_YOLO | false | Drop the OS sandbox for hook reviews (true/false, 1/0, on/off) |
bun install
bun run check # lint + format (ultracite)
bun run typecheck # tsc
bun run validate # validate the plugin and marketplace manifests
There is no automated test suite in this repo. Hook logic is exercised manually in live Claude Code sessions; Codex-backed reviews need auth and the network.
Inspect the hooks and skills inside your editor:
Claude Code:
/hooks
/correctionguy:correctionguy
Cursor:
Settings, Hooks
Settings, Plugins, correctionguy
/correctionguy
Before submitting, run bun run check, bun run typecheck, and bun run validate.
Submit the public repository at cursor.com/marketplace/publish. Licensed under MIT.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub correctionguy/correctionguy --plugin correctionguyHarness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
A growing collection of Claude-compatible academic workflow bundles. Covers scientific figures, manuscript writing and polishing, reviewer assessment, citation retrieval, data availability, paper reading, literature search, response letters, paper-to-PPTX conversion, and evidence-grounded Chinese invention patent drafting. Rules are organized as reusable skill folders with explicit workflows and quality checks.
Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.