Skill UI

Addresses the critical challenge of managing extremely long conversation histories (millions of tokens) in advanced AI agents, such as coding assistants. This technique moves beyond simple token minimization by implementing structured compression strategies (like Anchored Iterative Summarization) to prevent the loss of critical technical details—such as file paths, decisions, and error messages—ensuring the agent maintains functional context across complex, multi-step tasks. It emphasizes optimizing for 'tokens-per-task' rather than just tokens-per-request.

This skill implements a persistent, file-based planning system, inspired by Manus, allowing AI coding agents to store key project artifacts (task_plan.md, findings.md, progress.md) directly to disk. By externalizing the working memory, it ensures that complex, multi-step projects, research tasks, or coding efforts survive context window limitations, session restarts, or context loss. Essential for any deep, multi-phase workflow.

A powerful agentic workflow for complex software development. Cascade Write mode acts as an autonomous coding agent, allowing users to build features, refactor code across multiple files, run terminal commands, and debug errors—all from a single natural language prompt. Ideal for full-stack development, complex system building, and large-scale code refactoring.

This infrastructure provides a robust solution for running powerful AI agents in remote, sandboxed environments. It overcomes the limitations of local execution by ensuring unlimited concurrency, consistent environments, and collaborative features. Use this to build complex background coding agents, multi-client interfaces (e.g., Slack, Web), or scale agent workloads far beyond local machine constraints.

A robust framework for orchestrating complex, multi-step coding tasks using multiple AI agents. It functions by first planning the entire project into a mission dependency graph (DAG), then dispatching specialized agents to isolated Git worktrees for parallel development. The system automatically monitors progress, handles merges, and generates structured, comprehensive reports detailing changes, completed tasks, and next steps.

Blueprint turns a single objective into an executable, multi-step construction plan tailored for coding agents and multi-session teams, complete with context briefs, review gates, dependency awareness, and installable via `/blueprint <project> <objective>`.

Guides a project planning agent through researching, decomposing, and documenting multi-commit development plans before coding, enforcing autonomous research protocols and template-driven plan outputs for structured PR execution.

This guide details how to use sophisticated coding agents (such as Codex, Claude Code, or Pi) through a structured bash workflow. It provides methods for complex development tasks, including building new features, refactoring large codebases, and reviewing pull requests (PRs). Key best practices covered include managing pseudo-terminals (PTY), defining focused working directories (`workdir`), and executing tasks in the background for robust, multi-step automation.

Cavecrew provides a set of specialized subagents (investigator, builder, reviewer) designed for efficient code development delegation. Unlike vanilla agents, Cavecrew compresses output results, drastically reducing the token usage injected back into the main context. This allows users to maintain context over much longer, complex coding sessions, effectively preventing context exhaustion during large-scale refactoring, debugging, or multi-step feature development.

This skill generates comprehensive, tailored instruction files for AI coding assistants like GitHub Copilot and Claude. It supports multiple strategies (flat or nested) and is specifically designed for managing complex monorepos by creating area-scoped instructions using `applyTo` globs. Use it to standardize AI tooling usage across large codebases.

This skill implements advanced 'Gatekeeper' logic to optimize AI agent performance in long-running workflows. It prevents context window bloat by sharding metadata, allocating token budgets, and enforcing Atomic Precision to deliver purely functional, filler-free responses. Ideal for complex coding, multi-step reasoning, and large document analysis where conciseness is critical.

Open Dynamic Workflows (ODW) is an advanced, open-source engine designed to manage complex coding tasks using multiple AI agents. It operates in three stages: planning the task into parallel subtasks, orchestrating various coding agents (like Codex or Antigravity) to work concurrently, and finally, performing an adversarial verification pass to challenge and validate the combined output before merging. It supports multiple model providers and is ideal for structured, high-reliability software development.