Mini Agent

A 3P update is a succinct summary designed for executives and leadership to be read in 30-60 seconds. It covers a specific team's work over a set time period (typically one week).

Core Sections

1. Progress: Accomplishments from the past week (e.g., shipped features, milestones achieved, tasks completed).
2. Plans: High-priority objectives for the upcoming week.
3. Problems: Blockers or issues slowing the team down (e.g., resource shortages, bugs, failed deals).

Writing Guidelines

• Granularity: Adjust detail based on team size. Large teams (e.g., entire company) should focus on high-level impact (e.g., "hired 20 people"), while small teams can be more specific.
• Tone: Matter-of-fact and data-driven. Avoid heavy prose; use metrics where possible.
• Length: Each section must be strictly 1-3 sentences.
• Context: If the team name is not provided, ask for it before proceeding.

When building MCP (Model Context Protocol) servers, design tools specifically for AI agents rather than simple API wrappers. Follow these principles:

• Build for Workflows: Consolidate related operations into high-impact tools (e.g., a schedule_event tool that checks availability and creates the event) instead of exposing raw endpoints.
• Optimize for Limited Context: Return high-signal information. Provide options for "concise" vs "detailed" responses and prefer human-readable identifiers (names) over technical IDs.
• Design Actionable Error Messages: Errors should guide the agent. Instead of just diagnostic codes, suggest next steps like Try using filter='active_only' to reduce results.
• Natural Task Subdivisions: Use tool names that reflect human task thinking and group related tools with consistent prefixes.
• Use Evaluation-Driven Development: Create realistic evaluation scenarios early to let agent feedback drive tool improvements.

Define tool input schemas using Pydantic BaseModel. FastMCP uses these models to automatically generate inputSchema and perform validation.

Best Practices:

• Use ConfigDict for configuration (e.g., str_strip_whitespace=True, extra='forbid').
• Use Field to provide descriptions and constraints (e.g., min_length, ge, le).
• Use field_validator (with @classmethod) for custom validation logic.

from pydantic import BaseModel, Field, field_validator, ConfigDict
from typing import Optional, List

class ServiceToolInput(BaseModel):
    model_config = ConfigDict(
        str_strip_whitespace=True,
        validate_assignment=True,
        extra='forbid'
    )

    param1: str = Field(..., description="First parameter description", min_length=1, max_length=100)
    param2: Optional[int] = Field(default=None, description="Optional integer", ge=0, le=1000)
    tags: Optional[List[str]] = Field(default_factory=list, description="List of tags", max_items=10)

If the PDF is flat (non-fillable), you must manually define bounding boxes for text annotations. Follow these steps exactly:

1. Visual Analysis: Convert the PDF to PNGs to identify where data should go.

   python scripts/convert_pdf_to_images.py <file.pdf> <output_directory>

   Identify bounding boxes for both the label and the entry area. The label and entry boxes must not intersect. For checkboxes, the entry box should target only the small square, not the text label.

2. Create fields.json: Define the layout in a JSON file. Example fields.json structure:

   {
     "pages": [
       { "page_number": 1, "image_width": 1000, "image_height": 1400 }
     ],
     "form_fields": [
       {
         "page_number": 1,
         "description": "The user's last name",
         "field_label": "Last name",
         "label_bounding_box": [30, 125, 95, 142],
         "entry_bounding_box": [100, 125, 280, 142],
         "entry_text": { "text": "Johnson", "font_size": 14, "font_color": "000000" }
       }
     ]
   }

3. Generate and Validate Visuals:

   • Create validation images (red = entry, blue = label) for each page:

     python scripts/create_validation_image.py <page_number> <path_to_fields.json> <input_image_path> <output_image_path>

   • Run an automated intersection and height check:

     python scripts/check_bounding_boxes.py <JSON file>

   • CRITICAL: Manually inspect the validation images. Red rectangles must cover ONLY input areas and MUST NOT contain text. Blue rectangles should cover label text.

4. Apply Annotations: Create the final PDF.

   python scripts/fill_pdf_form_with_annotations.py <input_pdf_path> <path_to_fields.json> <output_pdf_path>