Commands

Reads Claude's replies out loud on macOS with ElevenLabs (via OpenRouter) or the built-in say voice. /speak to control it.

The record of orchestrated work: runs of units and tasks, the subagents spawned for them, each gate's verdict and the findings reviewers report, summed up in the status line while there is something to say. It can follow a superpowers plan document as its input. Run /ledger.

Prompt-cache meter above the Claude Code prompt: how many tokens each request read from, wrote to and sent past the cache, a live countdown to the cache's expiry, and what to do about it (keep going, compact or clear). The TTL follows Claude Code's own rules for your account (1 hour on a Claude subscription, 5 minutes on an API key, cloud provider or usage credits, plus the TTL variables and...

Learns from your corrections: keeps each one for you to confirm at session start, recalls the lesson when it is relevant again, and turns a repeated correction into a rule CLAUDE.md loads in every session.

Explains in Korean the shell command a permission dialog asks about, with what it would actually delete or discard (adapted from blast-radius), and asks before a risky loss in auto mode

See your LiteLLM virtual key inside Claude Code: budget, spend, limits, expiry, models and recent usage. Admins can also create and edit keys, grant extra budget and read router fallbacks.

A per-project todo list in a sidebar, kept in todos.md in the project folder, that both you and Claude can edit

Klondike solitaire in a pane, styled after the Microsoft Solitaire Collection: felt with a gradient, cards with shadows, block-letter ranks and pips, three difficulty levels, deals checked solvable by the built-in solver, hints, a solver that shows the solution, rescue (undo until solvable), auto-solve, demo, win animation. English or German.

Replay Theater: step through the file edits Claude made in the last turn, one diff at a time.

Kiko boxes every turn above your prompt: Knowledge In, Knowledge Out, K.O.!

Claude draws Mermaid, D2 and PlantUML diagrams (UML, flowcharts, sequences, architecture) into a side pane with history, versions, export, copy and share.

Desktop mod for people who run many sessions: a task brief with criteria and authority at the start, Claude works alone and calls you only for blockers, an acceptance screen and an HTML report at the end. A free mode for brainstorming and prototypes keeps the ideas and ends with a summary. The task folder keeps the ledger, so compaction loses nothing.

A right-hand sidebar for the session: every link mentioned, things to know, and action items you tick off into a collapsible Done list.

RUIN, a retro first-person shooter you play in a pane: four levels, textured walls, enemies, keys, weapons (/ruin)

Reports what a Claude Code session is really doing (idle / working / tool / permission, plus signals and token use) from inside the session, as a JSON snapshot and a JSONL event log

Shows your Obsidian Kanban board (BOARD.md) in the status line, with a /board pane listing every card

Keeps a running list of every file Claude edited or wrote this session: a count in the status line and a /touched pane

A band above the Claude Code prompt: an animated mini Claude that acts out what Claude is doing, who is working (model, effort, plan) and live ctx / 5h / 7d gauges.

A pane showing the main loop and every subagent of the session as pixel crabs, each under its role's hat: a short name, role, model, plan progress, context fill, tokens, estimated cost and time, with totals on top and the finished ones kept below; while agents work, one line above the prompt says how the run stands.

Draws successful tool calls as one dim line in the terminal and hides their output, keeping its key facts (git commit, background, exit code meaning); a call a hook refused is one red line; failed calls stay in full; ctrl+o shows everything; /quiet-tools toggles it.

Suggests next steps after every prompt and keeps them in a to-do list by category, in a panel you can click (/todo) or drive with commands.

Rewrites every fenced code block in an LLM response so it conforms to the language's standard style (Prettier for the web stack, black for Python, gofmt / rustfmt / shfmt where installed).

Read-only view of every user and assistant message in the current session, with collapsible tool-call details and paged loading. Open with /history.

Live plan progress bars with stages, steps, step times and soft sounds for decision, error and done. Local fork of zycck/claude-mods plan-progress 0.3.0 (ea2c96b) that sends its rules with the first prompt instead of prompt.compose and sends an unexplained open bar back from turn.complete instead of classic.Stop, redraws the bars from the transcript after a rewind or resume, folds the terminal...