AI agents are powerful. Now make them predictable.
Structured workflows that make every agent run repeatable. Scoped steps. Verified outputs. Full observability.
$ npm install -g @athenaflow/cliWorks with Claude Code and Codex today. More harnesses coming.
Install workflows
like packages.
Browse, install, and share community-built workflows. Each one is scoped, tested, and ready to run. Author your own or install from the marketplace.
playwright-automation
v0.0.21Writes, reviews, and stabilizes Playwright end-to-end tests from live product flows.
$ athena workflow install playwright-automationrobot-automation
v0.0.21Generates Robot Framework suites with keyword reuse, structured fixtures, and CI-ready runs.
$ athena workflow install robot-automationexploratory-testing
v0.0.14Runs charter-based exploratory sessions, captures evidence, and files actionable issues.
$ athena workflow install exploratory-testingsmoke-testing
v0.0.14Defines and executes smoke suites against critical paths so breakages surface before release.
$ athena workflow install smoke-testingregression-testing
v0.0.14Scopes impact-based regression runs, reruns the right areas, and reports deltas against the last release.
$ athena workflow install regression-testingThe agent is fine.
The workflow around it is broken.
- Same prompt. Same model. Different output every time.
- Agents drift over long runs. No checkpoints, no way to recover.
- What works for one dev doesn’t transfer to the team.
- You see the final output. Not the 40 tool calls that produced it.
- Agent tooling assumes a human at the keyboard. CI and automation are afterthoughts.
One workflow layer
across every harness.
AthenaFlow sits between you and the agent. Define steps, constraints, and exit criteria once — run on Claude Code, Codex, or any future harness. Same workflow, same result.
Install, run, ship.
Three steps.
Pick a workflow, run it, ship the result. Athena handles everything in between.
From marketplace or from scratch
Install a community workflow, point to a local file, or author one from scratch. One command to get started.
Run it anywhere. Athena handles the loop.
Interactive or headless. Athena manages permissions, checkpoints, and progress while the agent works.
Artifacts land where you need them
Every run is saved. Resume with full state. Pipe structured JSONL to files, CI artifacts, or downstream tools.
One layer across
every harness.
Each harness has its own way. Athena unifies them into a single workflow model.
Declarative workflows
One file defines prompts, plugins, isolation, and model preferences — portable across any harness. Version and share like code.
Real-time event feed
A unified observability layer. Every tool call, permission, and decision streams live — same format, any harness.
Unified session model
One session format across harnesses. Resume, inspect, and replay any run regardless of which agent executed it.
Harness-agnostic
One workflow definition. Runs identically on Claude Code, Codex, and future harnesses.
CI-native execution
One headless interface for any harness. JSONL output, structured exit codes, and permission policies for your pipeline.
Portable plugins
Each harness has its own plugin system. Athena provides a single interface to author and run plugins across all of them.
Start building.
Install the CLI, run the setup wizard, pick a workflow.
$ npm install -g @athenaflow/cli