claude-rudder

Generate a valid 'claude mcp add' command from a plain English description of what MCP server the user wants to install

Add new contextual information to the repository's context pool

Prune a bloated CLAUDE.md to its essentials and offload supplementary context into an agent-context/ directory.

Split CONTEXT.md into organized chunks in context-data/ directory

Extract information from CLAUDE.md and incorporate into CONTEXT.md

Run a structured health check on the current conversation context and report its quality score with actionable recommendations

Convert CONTEXT.md (human scratchpad) to CLAUDE.md (agent briefing)

List all available MCP servers in the catalog, optionally filtered by category

Manually edit and synchronize CONTEXT.md and CLAUDE.md

Show the full Claude Code MCP command syntax reference, including management commands, scopes, authentication types, and platform-specific notes

Remove outdated or incorrect information from the context pool

Evaluate whether the current conversation should continue or whether starting a new session would produce better results

Capture a follow-up task that surfaced mid-flow but shouldn't distract from the current focus. Appends to `<repo>/planning/leftover-tasks.md` (auto-scaffolds if missing). Use when the user says "park this", "leftover task", "note this for later", "follow-up but not now", "don't let me forget about X", or when Claude and the user mutually agree to defer something rather than context-switch.

Use when the user wants to prune and chunk a bloated CLAUDE.md into a structured context/ store with pipeline folders and subagent templates. Triggers on phrases like "chunk claude", "prune CLAUDE.md", "set up context store", "split CLAUDE.md".

Skill for evaluating whether a conversation has accumulated enough irrelevant context to warrant starting a new session. Applies structured criteria to distinguish valuable accumulated context from bloat that degrades response quality.

One-shot context-checkpoint workflow — when the current session is getting cluttered, write a structured handover plan capturing what's done and what's outstanding, then offer to spawn a fresh Claude session that opens already aware of the plan. Use when the user says "context is getting heavy", "let's checkpoint here", "stop, handover, restart", "context-gate", or otherwise signals they want to wrap the current session and continue cleanly in a new one.

Reference and worked example for the CONTEXT.md / CLAUDE.md two-file workflow. Use when the user asks "what is CONTEXT.md", how the human-context vs agent-briefing split works, or wants a copyable example pair to seed a new repo. Pairs with the operational skills (context-to-claude, claude-to-context, add-to-context, remove-from-context, manual-context-update, chunk-repo-context) — this skill explains the model; those skills do the work.

Use when the user wants to strip AI-generated style tics from documentation — emojis, excessive lists, all-caps filenames, fragmented structure. Triggers on phrases like "declaude the docs", "clean up documentation style", "remove emojis from docs", "apply style guide".

Synthesize a new SKILL.md from the procedure that was just followed in this session, so the next session can replay it without being briefed from scratch. Use when the user says "extract this as a skill", "save this process", "make a skill from what we just did", "this should be a skill", "let's not have to explain this again next time", or after a long worked example whose procedure has clear reuse value.

Create a new repo that starts from the current repo's state but takes a new direction — copy the code, seed CLAUDE.md with the revised approach, and push. Use when the user says "fork this in a new direction", "new repo from here with a new approach", or "branch this into a new project".

Use when the user points at a GitHub repo (an agent skill, MCP server, or "Claude plugin"-ish project) and wants to install or register it. Routes between native plugin install, MCP server registration, and adding to the user's third-party marketplace. Triggers on phrases like "install this from GitHub", "add this skill to my setup", "register this as a third-party plugin", "is this available as a Claude plugin".

Use when the user wants to install a new user-level slash command at ~/.claude/commands/. Triggers on phrases like "install a slash command", "add a /command", "create a user-level slash command", "save this as a slash command".

[EXPERIMENTAL] Use Agent Junction to relay context between two running Claude Code sessions in real time. Requires Agent Junction MCP server running on localhost or LAN. Use when the user wants two agents to talk to each other, share findings, or coordinate work across repos.

Append a known blocker to `<repo>/planning/blockers.md` (auto-scaffolds if missing). Use when something is preventing progress and shouldn't be lost — waiting on a code review, an external decision, an upstream bug, a credential not yet provisioned, a flaky test that needs investigation. Triggers on "log a blocker", "we're blocked on X", "note this blocker", "add to blockers".

Write a comprehensive point-in-time session log into `<repo>/planning/session-logs/YYYY-MM/YYYY-MM-DD-<slug>.md` for archival rather than immediate continuation. Captures everything achieved, the current overall state, decisions, surprises, and follow-ups for the distant future. Use when the user says "log this session", "wrap up with a record", "write today's session out", "record what we did", "we may not come back to this for a while — capture it", "session log", or otherwise signals they want a journal entry rather than a handover.

Teaches Claude the correct syntax for all Claude Code 'claude mcp add' command variants, covering HTTP, SSE, and stdio transports, scope options, authentication patterns, and command generation rules for MCP server installation.

Create a private GitHub repo that mirrors the current public repo. Use when the user says "mirror this to a private repo", "create a private mirror", or "private copy of this repo".

Canonical reference for where Claude Code plugins should persist user data. Defines the CLAUDE_USER_DATA convention, resolution order, and forbidden locations. Use this skill whenever designing or updating a plugin's onboarding flow, preferences storage, or persistent-state handling — and whenever Claude is tempted to write under ~/.claude/.

Create a new GitHub repo seeded with context from the current conversation. Use when the user says "repo from thread", "repo from chat", "let's turn this conversation into a repo", or "create a repo to manage this project" mid-conversation.

Save a Claude Code plan (from ExitPlanMode, design discussion, or multi-step task) to persistent storage. ONLY invoke when the user explicitly asks to save/stash/store a plan (e.g. "save this plan", "stash the plan", "store the plan in the planning repo"). Do NOT auto-invoke after planning. Routes to the active project's `planning/` folder when one exists, else to the user's configured central planning repo.

Create a lightweight `planning/` folder in the current repo with leftover-tasks.md, blockers.md, session-logs/, and handovers/. Idempotent — leaves existing files alone. Acknowledges task-queuer's heavyweight structure if present without duplicating. Use when the user says "scaffold planning", "set up planning folder", "init planning", or when another claude-rudder skill needs the structure and finds it missing.

First-run setup for the session log archive. Creates a private Agent-Logs repo on GitHub, clones it locally, and saves the path to memory so handovers auto-archive there. Triggers on phrases like "set up agent logs", "configure handover logging", "where should handovers go", "setup session logs".

Create a sibling docs repo for the current project, move documentation out of the main repo into it, seed CLAUDE.md, and push to GitHub. Use when the user says "spawn a docs repo", "split docs into its own repo", or "rip out the docs".

Show a one-screen cheat sheet of Claude Code primitives (skill, subagent, slash command, plugin, MCP server, hook, output style) and when to use each. Use when the user asks for a quick reference, an overview of the Claude Code extension surface, or "what are all the ways I can extend Claude Code".

Decide which Claude Code primitive (skill, subagent, slash command, plugin, MCP server, hook, output style) to use for a workflow idea. Triggers on "should I build this as a skill", "how do I extend Claude Code for X", "skill vs command vs plugin", "where should this live", or any question about choosing between Claude Code extension mechanisms.

Re-fetch the official Anthropic Claude Code documentation pages that this plugin grounds its recommendations in, and update reference/docs-snapshot.md with a fresh dated excerpt. Use when the user asks to refresh, update, or re-pull the Claude Code docs the plugin uses.

Capture the user's Claude Code development patterns (solo vs team, public vs private distribution, MCP comfort, default install scope) into the meta-tools plugin data directory so future recommendations from /what-thing:pick are tailored. Use when the user asks to set, update, review, or clear their what-thing context.

2 hooks across 2 events