By skyhook-io
Autonomous dev-loop (plan -> implement -> review -> PR -> converge), cross-model review, and product/UX critique as composable slash commands.
Autopilot — plan → implement → review → PR → converge, running the whole chain end-to-end and stopping only at real decisions
Run the native Codex review (codex plugin), print findings verbatim, then triage them skeptically
Cross-model review by the configured secondary reviewer (codex or cursor), printed verbatim, then triaged skeptically
Process the review findings from the conversation above and fix valid issues.
Autonomously iterate on the current PR until CI and automated reviewers converge, or until a human decision is needed.
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.
Composable AI dev-workflow commands for Claude Code — with Codex companions. An autonomous plan → implement → review → PR → converge loop, cross-model (Claude ↔ Codex) review, and a product/design/UX critique pass.
Built by Skyhook for building Radar; generic enough to use anywhere.
The loop
/autodev [--auto] <task> — full autopilot: plan → implement → review → PR → converge, end-to-end, stopping only at real decisions. Default mode gates on the plan before coding; --auto makes product calls itself (logging assumptions) and stops only for an irreversible-action ceiling./plan-loop — draft a plan, cross-review it with the other model, triage the critique skeptically, iterate, gate./review-loop — proactively review the current code (self + cross-model), triage, fix, update the PR — loop until clean.Review & critique (three altitudes — right thing → earning its keep → correct)
/product-review — questions what to build and how users perceive it: premise, top user journeys (ranked 0–10), UX shape, comprehension, states, AI-slop. Borrows forcing-function patterns from gstack and OneRedOak + a journey-ranking + non-expert-comprehension lens./review, /simple (anti-over-engineering), /triage-findings, /fix-findings.Cross-model review
/cross-review — runs a review by the configured secondary model — Codex or Cursor — prints it verbatim, then triages it skeptically (never auto-accepts). Pick the reviewer via ~/.claude/skyhook-skills.json ({"reviewer":"codex|cursor","model":"…"}), the SKYHOOK_REVIEWER env var, or a consult cursor / consult codex directive. /codex-review forces Codex.PR
/pr, /fix-pr, /fix-pr-loop (reacts to CI + bot reviewers until converged).Cross-cutting: never auto-accept a reviewer (your own, the cross-model pass, or PR bots) — every finding is triaged with evidence; cross-review only when nontrivial; every loop caps and reports; review at altitude (a clean implementation of the wrong thing is still wrong).
A few principles run through every command:
/product-review is the dedicated pass for this./triage-findings) and every loop that uses it inherits the change./review-loop consult codex, /autodev no PR, quick, focus on auth)./qa). The loops call /qa, but each repo defines what verification means there — type-check, tests, and optionally a /visual-test. The workflow stays repo-agnostic.Every loop narrates itself and closes with a scannable audit ledger:
📋 review-loop · feature/bulk-actions
🔗 PR [#910](https://github.com/your-org/your-repo/pull/910) · OPEN · checks 4 ✓ · 2 pending (Backend, Bugbot)
round 1
🔎 self-review 6 findings
🔵 codex review 8m · 13 findings
⚖️ triage 5 fix · 7 skip · 1 discuss
🔧 fix 5 applied
✅ qa tsc ✓ · test ✓ · visual-test skipped (no UI delta)
📤 PR pushed + body
result: converged locally · 1 round · 5 fixed · 7 skipped · 1 open · CI pending
open for you:
• [discuss] <the one question that needs your call>
Run these inside Claude Code (they're slash commands, not shell commands):
/plugin marketplace add skyhook-io/skyhook-skills
/plugin install skyhook-skills@skyhook-skills
The commands then appear namespaced under the plugin — /skyhook-skills:autodev, /skyhook-skills:review-loop, /skyhook-skills:product-review, etc. (/plugin to browse them). Auto-updates at startup — it's a public marketplace, no token needed.
Claude's marketplace only manages the Claude plugin; the Codex-side skills install with one line:
npx claudepluginhub skyhook-io/skyhook-skills --plugin skyhook-skillsComprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Harness-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 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.
Reliable automation, in-depth debugging, and performance analysis in Chrome using Chrome DevTools and Puppeteer
Feature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.
Real-time statusline HUD for Claude Code - context health, tool activity, agent tracking, and todo progress