Lollms Artifacts and Surgical Patching

Files (1)

• SKILL.md

SKILL.md (first lines)

# Lollms Artifacts and Surgical Patching

## 0. UI Header vs. Content Streaming Protocol
When artifacts are built or patched:
- **Header Presentation**: Only file names and structural boundaries (e.g., `Section: 1. Setup`, `def calculate_metrics()`, `class SessionManager`) are displayed in UI titles and subtitles.
- **Content Viewport**: Verbatim content, code lines, table records, and text streams exclusively inside the collapsible artifact container.
- **Immediate Real-time Flush**: Artifacts are synchronized to disk the exact millisecond the closing `</artifact>` tag is received, ensuring persistent safety across interruptions.

This skill explains how to utilize, create, and surgically update persistent artifacts using Lollms.

## 1. Creating and Updating Artifacts
Artifacts are persistent, versioned documents (code, text, notes) stored in discussion metadata. 
- Creating new: Provide 100% of the content inside `<artifact name="filename" type="type">...</artifact>`.
- Updating existing: Use **SEARCH/REPLACE Aider-style patches** to modify targeted regions.

## 2. Strategic Formulation: The Coding Plan
Before outputting any `<artifact>` update, the model must formulate a plan inside `<coding_plan>` to organize its implementation strategy:

```xml
<coding_plan>
  - Goal: Implement safe multiplication
  - Specialist Persona: Python Programmer
  - Target Artifacts: math_ops.py
  - Implementation Details: Add a method compute_multiply(a, b).
</coding_plan>
```

## 3. Surgical Patch Rules
When patching an existing file, the `SEARCH` block must copy lines **exactly as they appear** in the active artifact (matching all leading indentation spaces and comments character-for-character).

```xml
<artifact name="math_ops.py" type="code" language="python">
<<<<<<< SEARCH
        def compute_sum(a, b):
            # Target region
            return a + b
=======
        def compute_sum(a, b):
            # Target region
            return a + b

        def compute_product(a, b):
            # Safe multiplication logic
            return a * b
>>>>>>> REPLACE
</artifact>
```

## 4. Indentation-Agnostic Fuzzy Matching
If the LLM outputs a patch at column-0, or with slightly imperfect indentation, the Lollms Aider patch engine automatically:
1. Calculates the alignment offset of the matched region in the file.
2. Re-indents the `REPLACE` block to align with the rest of the code structure on disk.

## 5. Controlling In-Message Status Reporting
When performing complex tasks, the system can write status updates to the chat bubble. Use `enable_in_message_status` to toggle this behavior:
- `enable_in_message_status=True` (Default): Streaming chunks emit `<processing>` tags showing updates in real-time.
- `enable_in_message_status=False`: The prose stream remains 100% clean, and updates are fired only via background structured callbacks.
```

From ParisNeo/lollms_client, Apache-2.0.

More from ParisNeo/lollms_client