A file format for agent-friendly document navigation with generated indexes and structured section retrieval.
Status: active development. Breaking changes can occur.
- Avoid loading entire large docs for narrow questions.
- Provide stable section IDs and references (
{@id}) for traversal. - Keep INDEX and CONTENT synchronized via rebuild/validate workflows.
IATF stores both:
INDEX(generated navigation layer)CONTENT(source of truth you edit)
Typical flow:
- Edit
CONTENT. - Run
iatf rebuild <file>. - Run
iatf validate <file>. - Retrieve using
iatf index,iatf find,iatf read,iatf read-many,iatf graph.
Token efficiency is conditional and workload-dependent.
From docs/reports/agentic-usability-analysis.md:
- On sparse retrieval for a large fixture: ~64% fewer tokens vs full-file reads.
- On medium files with broad reads: index-first can be similar to full-file cost.
Use index-first flows when retrieval is selective and reference topology matters.
curl -fsSL https://raw.githubusercontent.com/Winds-AI/agent-traversal-file/main/installers/install.sh | sudo bashirm https://raw.githubusercontent.com/Winds-AI/agent-traversal-file/main/installers/install.ps1 | iexiatf rebuild examples/incident-playbook.iatf
iatf validate examples/incident-playbook.iatf
iatf index examples/incident-playbook.iatf
iatf find examples/incident-playbook.iatf "rollback"
iatf read examples/incident-playbook.iatf rollback
iatf read-many examples/incident-playbook.iatf detect rollback postmortemStart here: docs/README.md
Key docs:
docs/QUICKSTART.mddocs/COMMANDS.mddocs/SPECIFICATION.mddocs/DEVELOPMENT.mddocs/testing.mddocs/reports/agentic-usability-analysis.md
Archive (historical/exploratory):
docs/archive/IDEAS.mddocs/archive/TASKS.md
MIT License. See LICENSE.