Skip to main content

Attune AI v4.0: Your Workflows Now Have Their Own AI Teams

Patrick Roebuck
4 min read

15 workflows just got smarter. Each one now delegates to specialized Claude subagents that work in parallel.


The Big Idea

Every Attune AI workflow used to work the same way: send your code to one model, get a result back. It worked, but it meant a security audit and a code review used the same single-pass approach.

v4.0 changes that. Each workflow can now assemble its own team of Claude subagents — each one a specialist.

pip install 'attune-ai[developer]'

A security audit doesn't just scan anymore. It sends your code to a dependency checker, an injection detector, a secrets scanner, and a report writer — all running in parallel through the Anthropic Agent SDK.

What Changed

15 Agent SDK Adapters

Every major workflow now has an SDK-powered variant:

WorkflowWhat the Agents Do
security-auditDependency checker + injection detector + secrets scanner + report writer
code-reviewStyle analyzer + bug finder + architecture reviewer
test-genCoverage analyzer + test case designer + assertion writer
perf-auditHotspot profiler + complexity analyzer + optimization planner
bug-predictPattern matcher + history analyzer + risk scorer
doc-auditCoverage checker + style validator + link verifier
doc-genAPI extractor + example generator + formatter
health-checkDependency auditor + config validator + security checker
dependency-checkVersion scanner + vulnerability matcher + upgrade planner
refactor-planComplexity analyzer + dependency mapper + migration planner
release-prepSecurity auditor + test coverage checker + changelog validator + quality gate
research-synthesisSource gatherer + fact checker + synthesizer
simplify-codeRedundancy finder + abstraction flattener + dead code remover
test-auditGap finder + coverage analyzer + priority ranker

Each adapter uses 2-6 subagents depending on the task.

Smart Routing (Zero Config)

You don't pick between the old and new versions. Attune does it for you:

attune workflow run security-audit

If claude-agent-sdk is installed, you get the multi-agent version. If not, you get the API version. Same command, same output format. The only difference is depth and quality.

# Install the SDK to unlock multi-agent workflows
pip install claude-agent-sdk

Deduplicated Workflow Listing

attune workflow list now shows one entry per workflow with clear engine tags:

security-audit        [SDK]     Multi-agent security analysis
code-review           [SDK]     Parallel code review
perf-audit            [API]     Performance audit (SDK upgrade available)
test-gen              [native]  Test generation

No more duplicate entries. No confusion about which version to run.

CLI Tells You When an Upgrade Is Available

If you run a workflow that has an SDK variant but you don't have the SDK installed, the CLI tells you:

[security-audit] Running API version.
Tip: pip install claude-agent-sdk for multi-agent analysis.

One install command and your next run uses the full agent team.

Why This Matters

Single-model workflows hit a ceiling. One model can't be a security expert, a performance specialist, and a documentation reviewer at the same time. Multi-agent workflows break that ceiling by giving each concern its own specialist.

The result:

  • Deeper analysis — Each subagent focuses on one thing and does it well
  • Faster execution — Subagents run in parallel, not sequentially
  • Better reports — A dedicated report writer synthesizes findings from all specialists
  • Same cost — Smart tier routing still applies. Most subagents run on Haiku or Sonnet

What Else Shipped in v4.0

  • Bug fixes across 18 modules — Security, hooks, resilience, path validation, meta-workflow CLI
  • pip-audit CI workflow — Automated dependency vulnerability scanning in GitHub Actions
  • 207 new tests — Full coverage for all 13 SDK adapters plus behavioral tests for routing, dedup, and engine tagging
  • 11,800+ tests total at 80% coverage

Try It

# Install with agent support
pip install 'attune-ai[developer]'
pip install claude-agent-sdk

# Run a multi-agent security audit
attune workflow run security-audit --path ./src

# See all available workflows
attune workflow list

Or in Claude Code:

/attune security

Socratic discovery guides you through scoping, then the agent team handles the rest.


Links: PyPI | GitHub | Changelog

Related Articles