Guards

Claude as a small coworker in the terminal: an orange pixel-art Claude, two rows tall, in 132 frames. While a turn runs he sits beside the spinner in a scene for every spinner word (a thought bubble, a pen on a notepad, a book, a magnifying glass over text or over a globe, a laptop, a terminal, his own small helpers, a gear, a scroll for a skill, a plug for an MCP call); idle on the status row...

A small Claude avatar that runs back and forth above the prompt

Clawd in pixel art under the spinner while Claude works, with a twinkling progress backdrop

After Claude edits code, a side look at the task and the change shows one plain-language structural note above the prompt; it never enters Claude's context.

A small card above the prompt teaches you one thing while Claude works, and turns your day into something worth sharing

Commits files Claude edited as a git checkpoint after each turn, and adds /rollback to undo the last one.

Blocks reads of .env files and printing of API keys, with a red warning above the prompt.

/touched opens a pane listing every file Claude read or edited this session, with counts

Live token & USD observatory for Claude Code: per agent, subagent, workflow, task, skill and model, with prompt-cache countdowns, a model lock while the cache is warm, and subscription leverage

/incremental: the Claude mascot mines a rock for ore; every ore is tokens

See Claude's plan in a side pane while it writes it, with a comment button on every section

Shows the context tokens in use beside the prompt footer of the desktop app, green under 100k, yellow to 150k, red beyond. Needs function hooks (early access).

Pauses risky shell commands (rm -rf, force push, DROP TABLE) and asks you Proceed or Cancel first

A /edits command that opens a pane listing the files Claude changed this session, with Clear and Close buttons

A line above the prompt that says, in plain English, what Claude is doing right now and for how long

Explains each Claude Code permission request in plain English (WHAT, WHY, RISK) in a side pane

Agent hub: a pane listing this session's subagents with status, model, turns and tokens, and a box to steer a running one.

A persistent JavaScript kernel the model drives with one eval tool: state survives between cells, and cells call Claude Code tools, models and subagents.

oh-my-pi TTSR rules: a regex rule denies the tool call that would break it, with the rule as the reason; a question rule is judged after each turn.

An animated progress band above the prompt: one row per plan with its name, an LED-matrix bar whose pixels twinkle and brighten toward the head over a feathered glow, a lit chip riding the head that names the stage and the step under way ("Tests 3/5"), the percentage and a ✕ that dissolves the row. Claude reports its own progress through the report_progress tool it registers; a chime plays...

Shows the images you paste and the ones Claude reads, receives or creates in a small strip above the prompt: PNG, JPEG, GIF, WebP, HEIC, AVIF, TIFF, BMP and SVG.

Draws the context window as a stacked bar above the prompt, one colour per /context category. Toggle with /context-bar.

plan-progress fork with an animated rainbow pixel bar in the terminal, drawn under the prompt beside limit-bars: glide, shimmer, glow and a sparkle burst on done

Shows what Claude is running and adding to the model's context, the way pi does: every tool call on its own row, the end of Bash output, how long each call took, injected context, and a spinner naming the running call.