AI Workflow Orchestrator

A tiny CLI that runs Markdown-defined workflows using a simple DSL.

• Dry-run planning mode (emit actions)
• Real-mode execution via pluggable executors
• JSONL logging with start/end, status, duration

Quick start (local)

Clone and run the demo:

# Clone and enter repo
# git clone https://github.com/nam20485/ai-workflow-orchestrator.git
# cd ai-workflow-orchestrator

# Show help
node ./bin/orchestrator.mjs --help

# Run sample-minimal (real mode)
node ./bin/orchestrator.mjs \
  --workflow sample-minimal \
  --workflows-path ./examples/workflows \
  --exec-paths ./examples/executors \
  --params ./examples/params/sample-minimal.json \
  --console \
  --log logs/sample-minimal.jsonl

Logs are written to logs/*.jsonl. Artifacts are generated under logs/real-mode/ by the example executors.

Usage

ai-orchestrator --workflow sample-minimal --workflows-path ./examples/workflows --exec-paths ./examples/executors --console --log logs/run.jsonl

Key flags:

• --dry-run | --dry
• --workflow | --file
• --workflows-path
• --exec-paths <dir[,dir2;dir3]> (assignments/functions under these)
• --fixtures <file.json> (for dry-run function simulation)
• --params <file.json> (for real-mode params)
• --log <file.jsonl> (JSONL output)
• --console (log actions to stdout)
• --strict (treat warnings as errors in planning)
• --help (show help)

Package provides a bin named ai-orchestrator for global or npx use once published.

DSL

See ./docs/dynamic-workflow-syntax.md and ./docs/real-mode-plan.md.

CI

GitHub Actions runs a smoke test on each push: dry-run and real-mode for the sample workflow, and uploads the JSONL logs as artifacts.