Installable workflows for AI coding agents

Run your coding agent at the highest level

You've seen what's possible — agents that refactor incrementally, write tests first, review their own code, optimize the codebase for AI. But getting there means weeks of trial and error. Symposia is the shortcut.

~/your-project
$sym add sym-refactoring

+ sym-refactoring@1.0.0 — synced to 3 agents

$claude "refactor the auth module"

Verifying test coverage... 87%

Step 1/4: Extract validateToken()... tests green

Step 2/4: Move to auth/tokens.ts... tests green

Step 3/4: Rename legacy fields... tests green

Step 4/4: Remove dead code... tests green

4 commits, 0 regressions

$_
Works withClaude CodeCursorWindsurfGitHub CopilotAiderRoo CodeAmplify
Agent Capability Levels

Most teams are stuck at Level 2

Your agent generates code, but it doesn't engineer. No test-first discipline. No incremental refactoring. No self-review. The gap between Level 2 and Level 5 is the difference between a fast typist and a senior engineer.

Each Symposia package moves you up a level. The question is: where are you now?

See the path to Level 5
1

Chat

Answers questions, writes snippets. Forgets everything between sessions.

2

Generate

Writes features end-to-end. No tests-first, no incremental refactoring, no review. Fast but fragile.

3

Engineer

TDD with Red→Green→Refactor. Safe refactoring with test gates. Code review with severity triage. Process, not just output.

sym-tdd · sym-refactoring · superpowers-code-review

4

Orchestrate

Decomposes projects into phased plans. Dispatches parallel agents in isolated worktrees. Approval gates before execution.

sym-project-manager · superpowers-git-worktrees

5

Self-improve

Scores the codebase across 6 AI-readiness dimensions, generates fix plans, applies improvements. The agent optimizes its own environment.

adx

Why the gap exists

Getting from Level 2 to Level 5 isn't about better prompts. It's about teaching your agent multi-step engineering processes — and right now, there's no good way to do that.

You've seen what's possible

Demos of agents building full features autonomously. You try it and get spaghetti. The difference isn't the model — it's the workflow driving it.

The best engineers aren't prompting better

They've built multi-step processes — test-first, incremental refactoring, review gates. But that knowledge lives in their CLAUDE.md files, their custom setups, their heads.

There's no curriculum

Nobody teaches 'advanced AI-assisted engineering.' You're left reverse-engineering workflows from scattered blog posts and dotfiles. Trial and error, for weeks.

How Symposia works

Symposia packages are installable engineering workflows. Add one to your project, and your AI agent immediately follows the process — no configuration, no training.

1

Install

sym add sym-tdd writes structured workflow instructions into your project.

2

Agent reads

Your AI agent automatically discovers and follows the workflow — TDD, refactoring, review, whatever you installed.

3

Stack and compound

Add more packages. Your agent writes tests first, refactors safely, then reviews its own output — autonomously.

The path to Level 5

Each package unlocks a new capability. Start anywhere — they're standalone — but together they compound into an agent that engineers autonomously.

1
Level 3 — Engineer

Start writing tests first

Red → Green → Refactor cycle, enforced automatically. Your agent writes the failing test, implements the minimum to pass, then cleans up. Decision logic for when TDD applies and when to skip it.

sym add sym-tdd
2
Level 3 — Engineer

Refactor without breaking things

Atomic change → test → commit loop. Red triggers automatic revert. Behavior changes are forbidden during refactoring — the agent enforces the boundary. Every commit is a safe rollback point.

sym add sym-refactoring
3
Level 3 — Engineer

Review its own output

Your agent dispatches a reviewer subagent that evaluates plan alignment, code quality, and architecture — then categorizes issues as Critical, Important, or Minor with file:line references.

sym add superpowers-code-review
4
Level 4 — Orchestrate

Decompose and parallelize

Breaks projects into phased plans with explicit file ownership per task. Dispatches parallel agents in isolated worktrees. Progress tracked in PLAN.md — user approves before execution begins.

sym add sym-project-manager
5
Level 5 — Self-improve

Optimize the codebase for AI

6-dimension scoring — types, testing, guardrails, structure, context, readability — surfaces what's degrading agent output. Generates a prioritized fix plan and applies improvements automatically.

sym add adx
the compound effect
$sym add sym-tdd sym-refactoring superpowers-code-review

+ 3 packages synced to Claude Code, Cursor

$claude "add rate limiting to the API"

tdd

Writing failing test for rate limit middleware... red

Implementing minimum to pass... green

refactoring

Extract RateLimiter class... green

Move to middleware/rate-limit.ts... green

code review

Dispatching reviewer...

0 critical · 1 minor (naming) approved

3 workflows · 4 commits · 0 regressions

One natural language prompt. Three workflows fire automatically.

See it in action

These aren't guidelines your agent might follow. They're executable processes with verification gates, rollback points, and structured output.

Safe refactoring
sym add sym-refactoring

Incremental refactoring with automatic rollback

Your agent checks test coverage, then executes one atomic change at a time — extract, move, rename, delete — running the full test suite between each step. Red means instant revert, not a broken branch. Every commit is a safe rollback point.

Verifying test coverage... 87%

Extract validateToken() green

Move to auth/tokens.ts green

Rename legacy fields green

Remove dead code green

4 commits, 0 regressions

Codebase analysis
sym add adx

Make your codebase work for your agent, not against it

ADX runs a 6-dimension audit — types, testing, guardrails, structure, context, readability — and surfaces exactly what's degrading agent output. Then it generates a prioritized fix plan and applies improvements automatically, in severity order.

types
4/10
testing
7/10
context
2/10

23 escape hatches · 0 ADRs · no CLAUDE.md

run /adx-dashboard plan for fixes

Your best engineer figured out Level 4.
Everyone else is at Level 2.

The gap between your strongest AI user and everyone else keeps widening. Symposia makes advanced workflows installable — so the whole team climbs, not just one person.

One command onboarding

New hire runs sym add and gets every workflow the team uses — TDD, refactoring, code review — configured for every AI agent they have. Day-one productivity, not week-three.

Private org packages

Codify proprietary workflows — your deployment process, your review standards, your architecture patterns — as private packages only your org can install.

Create your own workflows

Distill hard-won engineering knowledge into packages with sym create-package. Your team's best practices become installable, versioned artifacts — not tribal knowledge.

Versioned and reviewable

Pin workflow versions across repos. Roll out updates deliberately. Review changes to agent instructions in PRs — the same way you review code.

Cross-agent consistency

One install syncs workflows to Claude Code, Cursor, Copilot, Aider, and every other agent on your team. No drift between tools, no per-agent configuration.

Publish to the community

Built a workflow that works? Publish it to the Symposia registry with sym publish. Share what you've learned with every team using AI agents.

Available now — open source

Start climbing

Three commands. Your agent goes from generating code to engineering it — with tests, safe refactoring, and self-review built in.

$ npm install -g symposia

$ sym init

$ sym add sym-tdd sym-refactoring superpowers-code-review

Stay in the loop

Get notified about new packages and features. No spam, unsubscribe anytime.