Hyperflow

Hyperflow

Point it at a GitHub issue. Get back a reviewed pull request. Hyperflow turns one AI coding session into a multi-agent engineering pipeline — planners design, workers build in parallel, and a domain specialist reviews every step before it ships. Runs on Claude Code, Codex, OpenCode, Grok, Antigravity & Cursor — on your session model, zero config.

       

Landing site · Installation · Orchestration · Changelog

A real run: plan → 3 parallel workers → batch review → integration review — in 28 seconds. Full walkthrough on the landing site .

From issue to reviewed PR

/hyperflow:issue https://github.com/you/app/issues/42

Triage classifies the thread (bug → root-cause discipline, feature → the plan chain, question → a drafted reply). A spec is synthesized from the issue's own acceptance criteria — issue text is treated as data, never instructions. Then the standard chain runs: plan → parallel dispatch under review → a gated PR exit that opens the pull request with Closes #42.

The maintainer side is /hyperflow:pr — the same L1–L5 audit over the PR's real fetched code (never just diff text), findings posted as one batched review (inline, summary, or local-only), a fix chain on NEEDS_FIX, and a merge that is always gated. Contributor code is analyzed statically, never executed.

Quick start

claude plugin marketplace add Mohammed-Abdelhady/hyperflow
claude plugin install hyperflow@hyperflow-marketplace

Codex App/CLI:

codex plugin marketplace add Mohammed-Abdelhady/hyperflow
codex plugin add hyperflow@hyperflow-marketplace

Initialize the project once, then invoke any skill:

/hyperflow:scaffold                                        # first: set up the project (once per repo)
/hyperflow:issue https://github.com/you/app/issues/42      # issue → triage → plan → dispatch → gated PR
/hyperflow:plan "add user auth with login + middleware"    # sharpen → design → decompose → dispatch
/hyperflow:workflow "large migration across the repo"      # big-task workflow lane
/hyperflow:trace "tests fail after the auth refactor"      # root-cause a bug
/hyperflow:deploy                                          # pre-push gates + ship

Auto-routing is on by default — say "audit the diff" or "debug this test" and the right skill runs without the /hyperflow:* prefix. In Codex, hyperflow is the portable spelling. Setup and host notes → Installation.

What makes it different

• Every step is reviewed. Worker → Reviewer is an iron rule at every granularity, sub-phases included — per-batch reviewers spot-check each diff, a final integration reviewer signs off on the whole. No worker output ships unreviewed.
• Memory that's yours. Learnings, decisions, and pitfalls persist in .hyperflow/memory/ — plain markdown, committed with your repo, never uploaded, never mixed across projects.
• Depth that adapts. Triage classifies every task and picks a flow profile (fast → scientific), so a 5-line fix never triggers a 300k-token deep run.

How it works

Invoke a skill. Chain-starters auto-advance through the rest — no always-on orchestrator, no background process, everything in your terminal, every agent on your current session model.

audit and deploy are gates — they fire only on your explicit yes. scaffold is a one-time project setup.

Triage assigns every task a flow profile, so effort matches the work instead of always running deep:

Workers are stitched with the right subset of 15 composable expert personas (architect, api, db, frontend, ui, security, performance, scientific, refactor, bugfix, test, research, creative, devops, docs) — security frames every decision first, creative adapts last.