Prompt

Give it a goal and success criteria; it plans, delegates, verifies and retries until the criteria pass or it needs you.

Flags typosquatted or brand-new packages before npm, pip or cargo installs them.

Starts your dev server in the background and shows its errors in a live pane.

One dashboard for every Claude session on your machine: what each is doing, cost, blockers, with pause and priorities.

The shared core that lets mods talk to each other: one event bus, one side panel with tabs, one place that routes notifications to your channels.

Blocks commands that upload files to external services like pastebins and file-sharing sites.

Claude proposes, you type: edits become diffs you apply yourself, for deliberate practice.

/explore sends three subagents to investigate different parts of the codebase and merges their findings.

A pane of your local database's tables and columns, ready to hand to Claude as context.

/replay steps through the session — prompts, tool calls, edits — in a pane, like a video timeline.

Routes every subagent to the right model for the task's difficulty, runs independent work in parallel and starts workflows when a job needs one.

A shared panel of team conventions, recommended mods, budgets and rules, synced through the repo.

A links bar above the prompt in Claude Desktop's Code tab: up to 4 pinned links per project or session with favicons, the session's 10 latest links, and a pane of every pinned link. Reads the session transcript; asks a link's own site for its page title and favicon (curl for PNG and ICO icons); opens links with open.

Painel lateral cyberpunk para o Claude Code (tokens/s, contexto, custo, agentes e ferramentas) e respostas renderizadas em neon: tabelas, codigo, diagramas, LaTeX e tool rows

Pasted images shown as a scrollable shelf above the prompt, even in iTerm2 + tmux; click one to mark it up

Spec Kit progress and usage governance for Claude Code: a phase rail above the prompt, the next command in the prompt hint, the current task on the spinner, a /astrolabe pane, toasts, update buttons, and a status entry with your usage window.

Routes each turn to the cheapest model + effort that fits the task (Haiku 5.5 → Sonnet 5.5 → Opus 5.5 → Fable 5.1) and shows the ladder in a side pane.

A band above the prompt with context size, per-turn cost and the prompt-cache countdown. Past the handoff line it counts down 60 s, writes a handoff file with a fork, runs /clear and resumes in the new conversation; a new conversation lists unclaimed handoff files, press 1 to resume.

Ambient language learning. Hooks weave vocabulary into ordinary Claude Code sessions — stale items you are forgetting, a small budget of new terms drawn from the work itself, and anything you use unprompted recorded rather than taught back at you. The ledger is a markdown file you own, written by a CLI rather than by hand.

A pane beside the session that Claude talks to you through: the task, what it is doing, the notes it files, the questions it went on without, and manual tests you run when you get to them.

Your Claude Code 5-hour and 7-day usage limits as animated bars under the prompt: percent, reset countdown, pace mark, projection, per-answer delta, styles (Pac-Man, beer, tank, battery, hourglass)

A web cockpit and a /tower pane over every Claude Code agent across your repos: see who needs you, approve and answer, send prompts and slash commands, start workstreams in git worktrees, and watch changes and PRs. Plus desktop notifications and a status line.

A pane showing every task and subagent as a colored progress bar with percentage and estimated time left

A clickable table of contents of the whole current session: topics grouped by category, with timestamps, in a pane