Agent Data Synthesis is an early-stage Python project for building an automated framework that generates, validates, and versions agent training data. The design target is not simple instruction-response expansion. The framework should synthesize executable agent trajectories across environments, tools, tasks, observations, and verification results.

• The repository currently contains the initial design documentation and a small local foundation runner.
• main.py builds a SQLite contact fixture, registers a typed lookup tool, executes candidate tasks, verifies trajectories independently, and writes JSONL plus a manifest.
• The architecture is documented before implementation so later code can follow stable domain boundaries.
• The current implementation is intentionally small; treat docs/ as the source of truth for design and development guidance.
• The planned synthesis pipeline is LLM-driven through a remote OpenAI-compatible API. It does not include local LLM cluster deployment.

• ARCHITECTURE.md: top-level system map.
• AGENTS.md: agent working guide and repository navigation.
• docs/README.md: canonical documentation index.
• docs/DESIGN.md: core architecture contracts.
• docs/design-docs/agent-data-synthesis-framework.md: deep design for the Agent data synthesis framework.
• docs/references/agent-data-synthesis-pdf-analysis.md: structured analysis of Agent-数据合成.pdf.
• docs/PLANS.md: implementation plan index.

uv run python main.py
uv run python main.py --output-dir artifacts/foundation --dataset-version dataset_foundation_v1
uv run python main.py --enable-refinement --output-dir artifacts/foundation-refined
uv run python main.py --enable-branching --output-dir artifacts/foundation-branching
uv run python main.py --enable-task-expansion --output-dir artifacts/foundation-task-expansion
uv run python main.py --enable-source-governance-fixture --output-dir artifacts/foundation-source-governance
uv run python main.py --use-llm --output-dir artifacts/foundation-llm
uv run python scripts/validate_docs.py
uv run python -m unittest

LLM-backed generation and judge steps should read these environment variables:

• AGENT_DATA_LLM_BASE_URL: OpenAI-compatible remote API base URL.
• AGENT_DATA_API_KEY: API key for the configured remote LLM provider.
• AGENT_DATA_LLM_MODEL: model id used by the synthesis pipeline.

The default local runner uses deterministic fixture candidates so it can run without provider credentials. Pass --enable-refinement to enable the deterministic one-shot critic/refinement fixture loop. Pass --enable-branching to include the deterministic multi-path branching fixture. Pass --enable-task-expansion to include deterministic seed transformation and task suggester/editor expansion. Pass --enable-source-governance-fixture to exercise deterministic no-network external-source governance and write source_events.jsonl. Pass --use-llm to generate candidates through the configured remote OpenAI-compatible /chat/completions API.

• Keep root files concise and use them as navigation entrypoints.
• Keep deep design, data, backend, security, and product decisions under docs/.
• Update docs and implementation in the same change whenever architecture, workflows, schemas, or entrypoints change.