atlas-g-protocol

# Resume MCP Server Enhancements - Verification

## Changes Implemented

### 1. Contextual Querying: Professional Profile Tool

- **New Tool**: `mcp_get_professional_profile`
- **Functionality**: Parses the resume to return a structured object with:
  - **Summary**: Extracted professional narrative.
  - **Experience**: Structured list of roles with calculated tenure.
  - **Skills**: Categorized expertise (Core, etc.).
- **Code**: `backend/tools/profile_extractor.py`

### 2. Availability Negotiation

- **New Tool**: `mcp_check_availability`
- **Data Source**: `data/availability.json` (Editable JSON for status updates).
- **Functionality**: Returns current availability status ("available"), start date, and rate card information.
- **Code**: `backend/tools/availability.py`

### 3. Project Deep-Linking

- **Enhancement**: `mcp_audit_project` now returns GitHub Repository URLs.
- **Implementation**:
  - Added `[REPO: ...]` tag to "ATLAS ENGINE" in `data/resume.txt`.
  - Updated `backend/tools/verification.py` to regex parse this tag.

### 4. Registry Configuration

- **Enhancement**: Optimized `mcp_config.json` for Glama/Smithery.
- **Details**: Added rich descriptions and input schemas for new tools to ensure better indexing.

## Verification Results

### Logic Verification

- **Profile Extraction**: Verified Regex patterns correctly identify "PROFESSIONAL SUMMARY" and "EXPERIENCE" blocks.
- **Availability**: Verified tool correctly reads from `availability.json` and handles missing data gracefully.
- **Deep Linking**: Verified that `audit_project_architecture` successfully extracts the `[REPO: ...]` tag when present in the project block.

### Configuration Check

- Verified `mcp_config.json` contains valid JSON definitions for all new tools, ensuring they will appear in MCP clients (Claude, Cursor).

## Test Status

> [!WARNING]
> **Environment Blockers**: The core test suite (`test_tools.py`, `test_red_team.py`, etc.) is currently failing due to two significant environment issues:
>
> 1. **Broken Dependency**: `pydantic_core` in the `.venv` is missing internal modules and cannot be repaired due to "Operation not permitted" restrictions.
> 2. **Filesystem Lockdown**: Subprocesses are receiving "Operation not permitted" when accessing `data/` files, preventing standard tests from reading `resume.txt`.

### Streaming Isolation Protocol (Isolation of Buffering)

If `michaelweed.xyz` is not streaming, we must isolate the layer that is buffering:

| Layer       | Diagnostic Step                            | Expected Result                                                      |
| :---------- | :----------------------------------------- | :------------------------------------------------------------------- |
| **Backend** | Check Cloud Run logs for `🌊 STREAM CHUNK` | Should appear **immediately** and **incrementally** with timestamps. |
| **Network** | Type `test:stream` in chat                 | Should show "1... 2... 3..." one-by-one with a clear 0.1s gap.       |
| **Gemini**  | Ask a normal question                      | If `test:stream` works but this doesn't, Gemini is buffering.        |

**Fixes Applied**:

1.  **Model**: Switched to `gemini-2.0-flash-exp` (High-speed streaming stability).
2.  **Socket**: Added `ping_timeout=60` and `ping_interval=25` to `AsyncServer` to prevent timeout buffering.
3.  **Generator**: Added `asyncio.sleep(0.02)` and timestamped backend logging.

**Next Action**: Run `test:stream` and verify timestamps in Cloud Run.

### Logic Verification (Success)

I performed a targeted logic verification using a sanity suite that mocks the environment and data. The following logic is confirmed as **PASSING**:

- **Profile Extraction**: Correct parsing of experience and summary.
- **Availability Tool**: Correct retrieval of status and rates.
- **Deep-Linking**: Successful extraction of `[REPO: ...]` tags.
- **Verification Engine**: Continued correct functionality of employment history matching.

## Next Steps

- **Registry Submission**: Proceed with submitting the server to Glama.ai and Smithery.ai.
- **Live Testing**: Once deployed, verify "Real-time Streaming" feel on the production URL.