feat: scaffold full atlas config from env init

[jbarnes850]

This pull request introduces several improvements to Atlas's environment discovery, configuration scaffolding, and LLM mocking capabilities. The main focus is on making the CLI more robust and user-friendly, especially for complex agent stacks and offline validation workflows. Key changes include enhanced .env loading, automatic Python path adjustments, provider/model autodiscovery, and the ability to run smoke tests with a fake LLM.

Environment and Configuration Discovery Improvements:

• The CLI now loads .env files from the project root and automatically adjusts PYTHONPATH to include both the project root and src directory, reducing manual setup for most stacks. (atlas/cli/run.py, atlas/utils/env.py, README.md) [1] [2] [3]
• Discovery auto-loads .env before spinning up factories and derives additional import hints from passed factories, streamlining agent/environment initialization. (README.md)

LLM Provider and Model Autodiscovery:

• Added recursive introspection logic to automatically detect LLM provider and model metadata from agent objects, supporting LangChain and custom stacks. This metadata is now included in the discovery payload. (atlas/sdk/discovery_worker.py) [1] [2] [3] [4]

LangChain Compatibility and Workarounds:

• Introduced a workaround to patch LangChain's pydantic helpers for better compatibility with wrapped tools and injected fields, preventing common errors in complex agent setups. (atlas/sdk/discovery_worker.py) [1] [2]

Offline Validation and LLM Mocking:

• Added support for the ATLAS_FAKE_LLM=1 environment variable, which enables mock LLM responses for offline validation and smoke testing without calling live providers. (atlas/utils/llm_client.py, README.md) [1] [2] [3] [4] [5]

Testing and Reliability:

• Added an integration test to verify that environment initialization scaffolds a runnable config, loads .env variables, and works with the mock LLM for smoke tests. (tests/integration/test_env_ready_config.py)

These changes collectively improve Atlas's developer experience, reliability, and support for advanced agent stacks and offline workflows.

[jbarnes850]

✅ Ready-to-run scaffolding overhaul is in. Highlights:

• Full config generator now builds clean Python adapter blocks and preserves template prompts/tools while injecting discovered LLM metadata.
• atlas env init auto-loads project .env, infers PYTHONPATH hints, and surfaces actionable failure guidance; the discovery worker now honors those overrides (and patches LangChain schema access).
• atlas run mirrors that bootstrap path so the generated config executes immediately.
• New integration test spins up a demo repo and exercises atlas env init --scaffold-config-full followed by atlas run under ATLAS_FAKE_LLM.

CI / smoke:

pytest
ATLAS_MOCK_AUTH0=1 ATLAS_FAKE_LLM=1 atlas env init --path external/auth0-assistant0/py-langchain/backend --env-fn examples.langgraph_adapter:create_environment --agent-fn examples.langgraph_adapter:create_langgraph_agent --env-arg graph=app.agents.assistant0:agent --agent-arg graph=app.agents.assistant0:agent --scaffold-config-full --force --no-run
ATLAS_FAKE_LLM=1 atlas run --path external/auth0-assistant0/py-langchain/backend --config external/auth0-assistant0/py-langchain/backend/.atlas/generated_config.yaml --task "Smoke task"

Let me know if you want additional coverage or docs tweaks.

[Copilot]

Pull Request Overview

This PR introduces a --scaffold-config-full flag for atlas env init that generates a complete, runnable Atlas configuration by merging template defaults with discovered factory metadata and inferred LLM provider/model information.

Key Changes:

• Auto-detection of LLM provider/model from LangChain-based agents during discovery
• Full YAML config scaffolding with discovery metadata, LLM parameters, and template defaults
• .env file auto-loading and intelligent PYTHONPATH inference during discovery
• Mock LLM mode (ATLAS_FAKE_LLM=1) for offline testing

Reviewed Changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/unit/test_env_scaffold.py	Unit tests for LLM metadata detection and full config composition logic
tests/integration/test_env_ready_config.py	Integration test verifying end-to-end scaffolding and config execution
atlas/sdk/discovery_worker.py	LLM metadata detection, LangChain workarounds, and environment/PYTHONPATH handling
atlas/cli/env.py	Full config scaffolding, auto-candidate selection, and .env loading logic
atlas/utils/llm_client.py	Mock LLM response implementation for offline testing
atlas/utils/env.py	Extended dotenv loader to accept optional path parameter
atlas/cli/run.py	Updated to load .env and configure sys.path
tests/unit/test_cli_run.py	Updated mock to accept variable arguments
README.md	Documentation for new scaffolding flag and features

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}