AI4Research V1

AI4Research V1 is a file-first multi-agent research production system. The initial P0 milestone focuses on a manually inspectable workflow:

brief -> planner task graph -> researcher artifacts -> reviewer gate -> writer input manifest

The system is intentionally simple at this stage. It uses local files, deterministic validation, and trace logs so that every research claim can be reviewed before it reaches the final report.

P0 Capabilities

• Initialize a run directory from a research brief.
• Create a complete seeded demo run.
• Validate task graphs, evidence cards, claims tables, and review reports.
• Enforce a deterministic review gate before artifacts are marked accepted.
• Build a writer input manifest from accepted artifacts only.
• Render the final Markdown report into a self-contained visual HTML report.
• Record run history in trace/events.jsonl.
• Run unit tests without external Python dependencies.

Quick Start

python3 -m orchestrator.cli init-run --brief docs/examples/example_brief.md
python3 -m orchestrator.cli demo-run
python3 -m orchestrator.cli validate --run runs/<run_id> --artifact plan/task_graph.json --kind task_graph
python3 -m orchestrator.cli gate --run runs/<run_id> --task T1 --version v1
python3 -m orchestrator.cli writer-manifest --run runs/<run_id>
python3 -m orchestrator.cli render-html --run runs/<run_id>
python3 -m unittest discover -s tests

Core Directories

docs/          Project plans, contracts, runbooks, and examples.
orchestrator/  P0 Python implementation.
schemas/       JSON Schema references for the artifact contracts.
templates/     Human-facing artifact templates.
.agents/       Codex skill skeletons for later P2 work.
runs/          Generated research runs.
tests/         Standard-library unit tests.

Current Phase

This repository is in P0. The goal is to prove the artifact and governance workflow before adding Codex MCP automation, external search tools, dashboards, or persistent databases.