Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By colliery-io
Persists and recalls conversation context across sessions using a memory palace, mines plan data from writes, and detects codex fallback on tasks.
npx claudepluginhub colliery-io/metisCancel active Metis Ralph loop
Explain Metis Ralph plugin and available commands
Decompose a Metis initiative into tasks
Execute all tasks under a decomposed initiative with Ralph loop
Execute a Metis task with Ralph loop
Use this agent to generate or update semantic summaries in .metis/code-index.md. Trigger after structural indexing (metis index) to add AI-generated understanding of what the code means — not just what symbols exist. Runs as a background subagent to avoid consuming main context. <example> Context: Code index was just regenerated and needs semantic summaries user: "Update the code index summaries" assistant: "I'll use the code-index-summarizer agent to generate semantic summaries." <commentary> Agent reads code-index.md, identifies modules needing summaries, reads ALL source files in each module, and writes semantic descriptions of symbols, relationships, and non-obvious behavior. </commentary> </example> <example> Context: PreCompact hook reports index was updated with changed directories assistant: "I'll use the code-index-summarizer to update summaries for changed modules." <commentary> Agent is given specific directories to update rather than re-doing everything. </commentary> </example>
Use this agent when working with Metis documents and needing methodology guidance. Helps with document type selection, work decomposition, phase transitions, and Flight Levels best practices. <example> Context: User wants to track some work but isn't sure what document type to use user: "I need to track this bug fix, what should I create?" assistant: "I'll use the flight-levels agent to help determine the right document type." <commentary> The agent provides guidance on document type selection based on the nature of the work. </commentary> </example> <example> Context: User is decomposing an initiative into tasks user: "Help me break down this initiative into tasks" assistant: "I'll use the flight-levels agent to guide the decomposition process." <commentary> The agent knows decomposition patterns and when/how to break work into tasks. </commentary> </example> <example> Context: User is unsure about phase transitions user: "When should I move this initiative to the active phase?" assistant: "I'll consult the flight-levels agent for phase transition guidance." <commentary> The agent understands exit criteria and phase flow for all document types. </commentary> </example>
This skill should be used when the user asks to "create a code index", "index this codebase", "update the code index", "generate code index", "build code index", "refresh semantic summaries", or needs guidance on generating or updating the .metis/code-index.md file for AI agent codebase navigation.
This skill should be used when the user asks to "break down this initiative", "decompose into tasks", "create tasks from initiative", "how to size tasks", "when to decompose", "vertical slices", "task granularity", or needs guidance on breaking higher-level work into lower-level work items.
This skill should be used when the user asks "what document type should I create", "create a bug ticket", "create a feature request", "should this be a task or initiative", "when to use an ADR", "track this bug", "log this tech debt", or needs help choosing between vision, strategy, initiative, task, backlog item, or ADR document types.
This skill should be used when the user asks "when to transition phases", "move to active", "exit criteria", "what phase comes next", "how to complete a task", "transition to completed", "mark as blocked", "phase flow", or needs guidance on advancing documents through their lifecycle phases.
This skill should be used when the user asks to "start a new project", "greenfield project", "tech debt campaign", "incident response", "feature development", "which preset", "initialize metis", "set up project", or needs guidance on project setup, choosing presets, and applying patterns for different work types.
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Uses Bash, Write, or Edit tools
Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Multi-project context management: switch projects, save/load sessions. Designed for non-CLI interfaces (Cowork desktop, Telegram).
AI-powered cascading development framework with design document system and multi-agent collaboration. Breaks down projects into Features (Mega Plan), Features into Stories (Hybrid Ralph), with auto-generated technical design docs, dependency-driven batch execution, Git Worktree isolation, and support for multiple AI agents (Codex, Amp, Aider, etc.).
Claude Code integration for MCP Task Orchestrator — schema-aware context, note-driven workflow
You work with me (Claude) - I guide your workflow and suggest next actions.
Qiushi Skill: methodology skills for AI agents guided by seeking truth from facts, with Claude Code, Cursor, OpenClaw, Codex, OpenCode, and Hermes guidance.
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Personal work and time management through reflection, transcripts, and sense-making. Provides MCP tools for capturing, extracting, and connecting the flow of work.
curl -fsSL https://raw.githubusercontent.com/colliery-io/metis/main/scripts/install.sh | bash
This installs the Metis desktop app. Launch it once -- on first launch it installs the metis CLI (which includes the MCP server) to your system PATH. After that, you're ready to connect Claude Code.
Manual download from GitHub Releases:
Metis_x.x.x_aarch64.dmgMetis_x.x.x_x64.dmgMetis_x.x.x_x64-setup.exeMetis_x.x.x_amd64.AppImagemacOS note: You may need to run sudo xattr -rd com.apple.quarantine "/Applications/Metis.app" after installing.
AI coding agents lose context constantly -- not just between sessions, but during them. Context compaction, token limits, and session boundaries all erase the plan, the progress, and the reasoning behind decisions already made.
Metis is a memory system for AI agents. It gives them a persistent, file-backed place to generate work, track progress, and record decisions -- so nothing is lost when context shifts. It uses the Flight Levels methodology to add opinionated structure around how work is created and tracked, so your AI doesn't just remember what to do, it works through it intentionally.
.metis/ directory. No lock-in, no proprietary formats. Read them in any editor, commit them with your code, diff them in PRs.After launching the Metis app once (so the CLI is installed), add the MCP server and plugin:
claude mcp add --scope user --transport stdio metis -- metis mcp
Then inside Claude Code:
/plugin marketplace add colliery-io/metis
/plugin install metis@colliery-io-metis
With the plugin installed, Metis detects your project automatically. Just talk to Claude:
You: "Initialize Metis for this project."
Claude creates the .metis/ directory and works with you to write a vision statement -- the purpose and direction of the project. Even for an existing codebase, the vision captures where the project is headed so all future work traces back to it.
You: "I want to add real-time collaboration to this app."
Claude creates initiatives under the vision. You discuss scope, approve the direction, and shape what gets built:
"Build a real-time collaboration platform" # Vision - the why
├── "Implement WebSocket infrastructure" # Initiative - a project
├── "Build presence system" # Initiative
└── "Design conflict resolution" # Initiative
You: "Break down the WebSocket initiative into tasks."
Claude proposes a task breakdown, you review and adjust, and Claude creates the tasks:
"Implement WebSocket infrastructure"
├── "Set up WebSocket server" # Task
├── "Add connection pooling" # Task
└── "Write integration tests" # Task
You: "/metis-ralph PROJ-T-0001"
Claude picks up the task, transitions it to active, and works on it autonomously -- writing code, running tests, updating the task document with progress. When it's done, it signals for your review.
This is the key idea. While a task is active, the AI continuously updates it with:
If the session ends or context compacts, the next session reads the task and picks up seamlessly. No work is lost.
Choose the right level of structure for your project: