aiwf

AI Workflow Framework for repeatable, logged, multi-stage generation pipelines.

Scope

aiwf is a framework to run AI workflows with:

• provider abstraction (OpenAI-compatible APIs)
• workflow runners (retries, batching)
• session logging and artifact capture
• plugins for domain tools (e.g., FlowMark)

Usage

aiwf tool add flowmark --cmd "flowmark"
aiwf run --tool flowmark -- validate docs/dev/v0.1.2/samples/minimal.md

With spec stack:

aiwf run --tool flowmark --spec-stack v0.1,v0.1.1,v0.1.2 -- validate docs/dev/v0.1.2/samples/minimal.md

With participants:

aiwf run --tool flowmark --participant "llm:responder" --participant "human:actor" -- validate docs/dev/v0.1.2/samples/minimal.md

List tools:

aiwf tool list

This creates a session directory under ~/.aiwf/sessions with:

• run.json
• events.jsonl
• artifacts/stdout.txt
• artifacts/stderr.txt

See docs/sessions.md for details.

See docs/tools.md for the tools registry format.

v0.1.3 CLI (partial)

aiwf session new
aiwf session list --format json
aiwf material add ./docs/spec.md --type file
aiwf material set create --include material_20260212_120000_abcd
aiwf catalog materials --format table
aiwf fetch material:material_20260212_120000_abcd --format json --pretty

Exit Codes

• 0 : command succeeded
• 1+ : command failed (propagates underlying command exit code)

Relationship to FlowMark

FlowMark remains a document format. aiwf provides an execution layer for generation, validation, and review pipelines.

License

MIT

npm pack (Publish Check)

npm pack
tar -tf hak2i-aiwf-0.1.2.tgz

Check that bin/, src/, docs/, README.md, and LICENSE are included.