A specialized tool to manage context isolation and focused task execution with Claude Code, solving the critical challenge of context length limitations and task focus when working with Claude on complex, multi-step projects.
Claude Task Manager solves a fundamental challenge when working with Large Language Models like Claude on complex projects: context length limitations and maintaining focus on the current task.
Getting the setup right took many iterations because you were wrestling with experimental features (vector indexes in 3.12.4), driver limitations (python-arango doesn’t expose “hidden” indexes), API mismatches (ArangoSearch view properties versus strict dict comparison), and fragmented documentation spread across release notes, HTTP API docs, driver specs, and blog posts. Each step revealed a nuance—from needing withHidden=true on an HTTP call to ignoring server-injected default view‐link properties—that isn’t obvious without deep dives into multiple docs (Features and Improvements in ArangoDB 3.12, API Changes in ArangoDB 3.12).
vector index type is not part of the cor
| { | |
| "customModes": [ | |
| { | |
| "slug": "planner", | |
| "name": "📝 Planner", | |
| "roleDefinition": "You are Roo, an experienced technical planner managing `task.md`. You identify major tasks/phases, delegate each exclusively to Boomerang Mode, process the overall results from Boomerang, handle final Git actions, log planning-specific lessons, and escalate unresolvable issues.", | |
| "customInstructions": "Your primary goal is to drive the project forward by managing the `task.md` plan phase by phase.\n\n1. **Identify Next Task:** Read `task.md` and find the first task marked with `[ ]`.\n2. **Delegate to Orchestrator:** Use the `new_task` tool to delegate the **entire identified task** (e.g., 'Task 8.3: Execute End-to-End Tests') and its description/sub-actions to `Boomerang Mode`, providing the full context.\n3. **Handle Task Completion:** When `Boomerang Mode` reports overall success for the delegated task via `attempt_completion`:\n a. Review the confirmation message.\n b. If satisfactory, p |
| #!/bin/bash | |
| # Required parameters: | |
| # @raycast.schemaVersion 1 | |
| # @raycast.title Repomix Copy | |
| # @raycast.mode compact | |
| # @raycast.packageName Repomix Tools | |
| # @raycast.icon clipboard | |
| # @raycast.description Pack a GitHub repository into clipboard using Repomix | |
| # @raycast.author Graham Anderson |
| { | |
| "rules": [ | |
| { | |
| "type": "code_advice", | |
| "condition": "contains_deprecated_methods", | |
| "action": "verify_against_documentation", | |
| "message": "Never suggest code without verifying each method/function against official documentation to ensure methods are not deprecated. Always provide a link to the specific documentation. If you do not have a link, do not suggest a solution. Instead, ask the user to provide a documentation link, or, ask the user what you need to provide a solution." | |
| }, | |
| { | |
| "type": "documentation_first", |
A specialized tool designed for AI agents to autonomously analyze Python packages, discover existing methods, and validate APIs before implementing new solutions. This tool helps prevent redundant code creation by identifying existing functionality in non-standard packages.
MCP's Python and TypeScript SDKs fail to run persistently in Docker. Despite setting a lifespan in the Python SDK, mcp.run() exits immediately, shutting down the container. This issue affects all tested MCP projects, making real-world deployment impossible.
This README explains how a Latent Program Networks (LPN)-inspired approach can generate tool execution strategies for multi-tool tasks in an agent workflow. This method offers advantages over relying solely on Large Language Models (LLMs) for tool strategy via prompting. A toy Python script demonstrates the concept in action.
Latent Program Networks (LPNs), originally designed for abstract reasoning (e.g., the ARC benchmark), embed "rules" or "completions" into a searchable latent space. Here, we adapt LPNs to embed tool configurations and execution order, enabling smooth coordination of multiple tools (e.g., generate_summary and create_pie_chart) for tasks like summarization and data visualization.
| #!/usr/bin/env python3 | |
| """ | |
| Dynamic and Adaptive Python Environment in a Bubblewrap Sandbox | |
| Overview: | |
| This project demonstrates a minimal, safe, and self-contained Python execution | |
| environment using bubblewrap (bwrap) and uv. The goal is to provide a lightweight | |
| alternative to Docker for running agent code—in this case, dynamically generated | |
| Python code along with its dependencies—within an isolated sandbox. |