Package Discovery Tracking

Relevant source files

The following files were used as context for generating this wiki page:

• models.py
• session_manager.py
• mcp_tools.py
• tests/test_session_manager.py
• tests/test_models.py
• tests/test_mcp_tools.py

Package Discovery Tracking identifies, suggests, and records software libraries relevant to specific development tasks within the Sequential Thinking MCP environment. This system helps you determine which tools are already installed or should be acquired to solve the problem defined in your session.

The feature integrates with the session lifecycle, triggering during session initialization and thought addition to ensure that architectural decisions remain grounded in available or recommended technical assets.

Sources: session_manager.py:53-76, models.py

Package Data Model

The system represents every discovered library as a PackageInfo object. This structure maintains metadata used for relevance ranking and architectural documentation.

PackageInfo Fields

Field	Type	Description
name	str	The unique identifier of the package.
version	str	The specific version string of the library.
description	str	A brief summary of the package's purpose.
relevance_score	float	A value from 0.0 to 1.0 indicating task alignment.
installation_status	str	Current state (e.g., "installed").
api_signatures	List[str]	List of key functions or classes provided by the package.
discovered_at	datetime	Timestamp of when the package was first identified.

Sources: models.py, tests/test_models.py:200-230

Relationship Diagram

This diagram shows how PackageInfo relates to the UnifiedSession and other session components.

Discovery and Exploration Logic

The discovery process runs through the UnifiedSessionManager.explore_packages method. It scans the environment's metadata and compares package information against the task description.

Discovery Workflow

1. Scanning: The system accesses importlib.metadata.distributions() to retrieve all packages installed in the current environment.
2. Scoring: It invokes _calculate_relevance to compare the package name and description against the task_description.
3. Filtering: Packages with a relevance score exceeding 0.3 are added to the session's discovered_packages list.
4. Persistence: The manager saves the updated list to the session's Markdown file under the "## Discovered Packages" heading.

Sources: session_manager.py:284-307, session_manager.py:126-136

Relevance and Keyword Mapping

The system utilizes a internal keyword map to suggest packages even if they are not currently installed. This helps you identify necessary tools during the planning phase.

Relevance Calculation

The system calculates relevance using basic string matching:

• Exact Match: A package name appearing directly in the task description receives a score of 0.8.
• Partial Match: Common words shared between the task and package name result in a score of 0.5.
• Default: Unrelated packages receive a baseline score of 0.1.

Sources: session_manager.py:401-409

Keyword Suggestions

When you add a thought with explore_packages=True, the system checks for specific keywords in your text to suggest relevant libraries.

Keyword	Suggested Packages
http	requests, httpx, urllib3
web	flask, django, fastapi
data	pandas, numpy, sqlite3
test	pytest, unittest, mock
json	json, jsonschema, pydantic
async	asyncio, aiohttp, tornado

Sources: session_manager.py:388-397

MCP Tool Integration

The MCPToolsHandler exposes package discovery through specific tool endpoints, allowing the AI agent to trigger exploration manually or automatically.

Tool: explore_packages

This tool performs a deep scan of the environment based on a provided description.

Automatic Exploration

Automatic discovery occurs under two conditions:

1. Session Start: If session_type is coding and package_exploration_required is True.
2. Thought Addition: If the add_thought call includes the explore_packages=True argument.

Sources: session_manager.py:74-76, session_manager.py:186-193

Summary of Capabilities

Package Discovery Tracking serves as the bridge between abstract reasoning and concrete implementation. By scanning installed distributions and mapping keywords to known libraries, the system ensures that you are aware of the technical resources available to solve a specific problem. All discovered data is persisted in the memory-bank/sessions/ directory, providing a permanent record of the technical environment for every session.

Sources: session_manager.py:126-136, README.md