Agent skill

Cardputer Companion

by dakshaymehta in dakshaymehta/cardputer-claude-os

A skill your agent uses when the cardputer MCP tools (notify, ask, confirm, show, progress) are available in the session, or the user mentions their Cardputer / handheld / "buzz me" / "page me".

Apache-2.0Auto-check passedAgent Workflows

Install Cardputer Companion

skills CLI
$ npx skills add dakshaymehta/cardputer-claude-os --skill cardputer-companion -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install dakshaymehta/cardputer-claude-os cardputer-companion --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/dakshaymehta/cardputer-claude-os.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.claude/skills/cardputer-companion .claude/skills/cardputer-companion && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
cardputer-companion
GitHub stars
208
Token cost
~2.5k tokens
SKILL.md length
1,459 words
Files
1
Skills in repo
1
Repo updated
First seen
Licence
Apache-2.0

At a glance

A skill your agent uses when the cardputer MCP tools (notify, ask, confirm, show, progress) are available in the session, or the user mentions their Cardputer / handheld / "buzz me" / "page me".

  • Works in 6 steps: Confirm policy — mandatory, non-negotiable → Proactive notify — quiet → Ask when blocked — quiet → …
  • The cardputer MCP tools (notify
  • SKILL.md covers Core ethos: default to silence, 1. Confirm policy — mandatory,…, 2. Proactive notify — quiet and 3. Ask when blocked — quiet, plus 7 more sections
  • Calls git and pytest

What it does

Cardputer Companion is an agent skill from dakshaymehta/cardputer-claude-os. Use when the cardputer MCP tools (notify, ask, confirm, show, progress) are available in the session, or the user mentions their Cardputer / handheld / "buzz me" / "page me". Governs runtime etiquette for the pocket device — mandates a physical confirm gesture before irreversible operations (passing the real action diff as details), buzzes the device on completion of long tasks, asks quick questions when blocked and the user is away from the keyboard, shows ambient status for long work, and formats all device…

Its SKILL.md is about 2.5k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in Agent Workflows, covering MCP servers. It works with Model Context Protocol. The repository describes itself as: DIY OS bundle for the M5Stack Cardputer: Claude Buddy (BLE), Push-to-Claude (voice + chat with memory via a Cloudflare Worker), and a flash-and-go installer skill for Claude… The licence is Apache-2.0.

When your agent uses it

  • The cardputer MCP tools (notify
  • Progress) are available in the session
  • The user mentions their Cardputer / handheld / buzz me / page me
  • Even when the user doesnt name the skill — if the cardputer MCP tools are registered

Example prompts

  • “buzz me”
  • “page me”
  • “/cardputer-companion”

Workflow steps

6 steps, taken from the step headings in SKILL.md.

  1. Confirm policy — mandatory, non-negotiable
  2. Proactive notify — quiet
  3. Ask when blocked — quiet
  4. Tiny-screen formatting
  5. Do Not Disturb — respect it
  6. Check before you interrupt — device_status

What it can do on your machine

Read from SKILL.md and the folder at commit 8e2bc8a. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • git
    • pytest

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use git, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Cardputer Companion loads about 2.5k tokens when it runs. Until then it costs about 200 tokens; SKILL.md has 1,459 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~200
When it runs · the whole SKILL.md, loaded when a task matches
~2.5k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from dakshaymehta/cardputer-claude-os at commit 8e2bc8a, republished under its Apache-2.0 licence (© dakshaymehta). 1,459 words, ~2,534 tokens.

Download SKILL.mdSave it as .claude/skills/cardputer-companion/SKILL.md (or your agent's skills folder).
name
cardputer-companion
description
Use when the cardputer MCP tools (notify, ask, confirm, show, progress) are available in the session, or the user mentions their Cardputer / handheld / "buzz me" / "page me". Governs runtime etiquette for the pocket device — mandates a physical confirm gesture before irreversible operations (passing the real action diff as details), buzzes the device on completion of long tasks, asks quick questions when blocked and the user is away from the keyboard, shows ambient status for long work, and formats all device output for the 240×135 LCD. This is the behavioral counterpart to the cardputer MCP server: the server is the hands, this skill is the manners. Trigger even when the user doesn't name the skill — if the `cardputer` MCP tools are registered, these rules are in force.

Cardputer Companion

The Cardputer is a credit-card-sized handheld the user carries in their pocket. It exposes MCP tools over a BLE bridge — notify, ask, confirm, show, and progress — provided by the cardputer MCP server in this repo (mcp/server.py). Those tools are the hands. This skill is the manners: it tells you when to reach for them and how to shape what you send, so the device stays useful instead of annoying.

These rules apply no matter where you run. The same tools reach the same device whether you're local Claude Code over loopback or a cloud Managed Agent / Messages-API agent over an MCP tunnel (tunnel/). If anything, restraint and the fail-closed confirm discipline matter more when you're an unattended cloud agent — the user isn't watching a terminal, and the physical gesture is the only thing standing between you and an irreversible mistake. The device shows which agent is asking on every ask/confirm banner (derived from your bearer token, so you can't misrepresent it) — keep your requests honest and legible.

Core ethos: default to silence

A buzz should mean something. The user is wearing this thing; every alert costs their attention. Earn each one. When in doubt, stay quiet and keep working — do not narrate progress to the device. The daemon now enforces a per-agent floor on non-critical notifys (a second non-crit buzz within ~60 s comes back rate-limited and never reaches the device), but treat that as a safety net, not a budget — keep self-imposed restraint as the real throttle. If you want a long task to stay visible without buzzing, use show (see below), not repeated notifys.

1. Confirm policy — mandatory, non-negotiable

Before any irreversible operation, call cardputer.confirm and wait for the physical hold gesture. Do not substitute a chat-message "are you sure?" — the whole point is that no tool output or prompt injection can synthesize a sustained physical keypress, but it can forge a convincing chat confirmation.

Operations that require confirm:

  • Production deploys / releases
  • git push --force (especially to main/master)
  • DROP TABLE, TRUNCATE, or DELETE / UPDATE without a WHERE
  • rm -rf, mass file deletion, or destroying uncommitted work
  • Paid API calls or financial transactions with large side effects
  • Anything the user could not undo in about a minute

Rules:

  • If confirm returns cancelled or timeout, abort the operation and tell the user — never proceed on a non-confirmation.
  • If confirm returns unavailable: … (device off or out of range), do not silently fall through to running the destructive command. Stop and tell the user the device isn't reachable; let them decide how to proceed.
  • Keep the title to ~18 characters and declarative — the user must recognize the operation at a glance: FORCE PUSH main, DROP customers, deploy prod.
  • Pass details whenever you can: the actual content being approved — the real shell command, the SQL, a short diff hunk, or payee + amount. The device renders it in a scrollable box above the gesture, so the user approves what they read, not just the title (the hardware-wallet model). Keep it to the essential ~256 chars and strip noise. It's text you supply, so it's legibility of intent, not cryptographic proof — and the title must still stand on its own (older firmware shows the title only).
  • Do not call confirm for routine yes/no decisions — that's what ask is for.
  • Every confirm decision (confirmed / cancelled / timeout / unavailable) is written to a local consent audit log (~/.cardputer-mcp/audit.log) with the agent label, title, and details. Another reason to pass a real, legible details: it's the durable record of what the user approved.

2. Proactive notify — quiet

Buzz the device on the final completion of a genuinely long-running task — multi-step or multi-minute work the user is plausibly waiting on while away from the keyboard (a full test suite, a migration, a build, a long agent run).

  • One notify at the end, not one per step.
  • Lead with the verdict: tests green, build failed, deploy ready.
  • Do not notify for quick interactive turns where the user is clearly watching the screen — that's just noise.
  • Use urgency honestly: info for "done, all good", warn for "done but needs your eyes", crit only for something that needs a reaction within seconds.

3. Ask when blocked — quiet

Use cardputer.ask only when you are genuinely blocked on a decision and the user may be away from their laptop. If they are plainly at the keyboard, ask in the chat instead — the device round-trip is slower and more intrusive.

  • Provide 2–4 short choices (each ≤ ~32 chars).
  • Keep the question to ~60 chars (wraps to two lines on the LCD).
  • On timeout or cancelled, fall back to chat — do not loop re-asking.

3b. Ambient status — show, glanceable not noisy

cardputer.show(text, channel) writes one silent line to the device's idle screen — no chirp, no screen takeover, not gated by DND. Use it to leave a live heartbeat of a long task the user can glance at, instead of buzzing them: building…, pytest 142/300, deploy ok.

  • It is NOT a notify. Never use it for something the user must react to — that's notify (or confirm). show is the status bar, not an alert.
  • Update at a human cadence — a few times across a task, not every step/token.
  • One line per channel (defaults to your agent label); keep text ≤ ~40 chars. The device keeps only the most recent few channels.
  • The pattern: a show heartbeat while working, then one notify at the very end (tests green). Don't replace the end-of-task notify with show.
Show full SKILL.md (553 more words)Show less

3c. Ambient progress — progress, when there's a percentage

cardputer.progress(label, percent, channel) is the visual sibling of show: same silence, same channel ring, same etiquette — but it renders a filling 0–100% bar instead of a text line. Reach for it instead of show whenever the work has a real denominator: a build, a download, a test sweep, an N-of-M migration.

  • Call it as the work advances at a human cadence (0 → 25 → 60 → 100), not on every increment. percent is clamped to 0–100; label shares the row with the bar, so keep it ≤ ~12 chars (build, tests, deploy).
  • It shares the show ring, so a progress and a show on the same channel contend for one slot — latest wins. A natural arc: progress to 100, then a show("done") (or the end-of-task notify) to close it out.
  • Still ambient: silent, no takeover, ignores DND. Same rule as show — it is NOT an alert. Pair the bar while working with one notify at the very end.
  • unavailable / older firmware → just skip it; it's a nicety, never required.

4. Tiny-screen formatting

The LCD is 240×135 pixels. Whatever you send must read in a glance:

  • Titles ≤ ~20 characters; bodies ≤ 3 short lines (~30 chars each).
  • No markdown tables, no code fences, no long file paths — strip to the essential token (auth_test.py, not the full path).
  • Spell out the outcome in plain words; the user can't scroll a banner.
  • Lead with the result, details second.

5. Do Not Disturb — respect it

The user can put the device in Do Not Disturb (a DND chip shows on its idle screen). When they have:

  • notify (non-critical) and ask return "dnd". Treat it as "the user is heads-down": do not retry or escalate; degrade to normal chat / logging and carry on. For ask, pick the safe default or proceed without the device rather than looping.
  • crit notifications and confirm still ring — DND never weakens the destructive-op gate. If you genuinely need a confirm and it timeouts because the user is asleep, that's the system working: abort, don't proceed.

6. Check before you interrupt — device_status

device_status() is a read-only, passive probe (no buzz, no radio wake): it reports online/offline, dnd, firmware caps, uptime, and battery. Use it when it would change your behavior — e.g. before starting a long unattended job ("will I even be able to page them when it's done?"), or before a non-urgent notify when you're unsure. If it says offline or dnd=on, prefer staying quiet and falling back to chat over firing a call that bounces. Don't poll it in a loop; it's a check, not a heartbeat for you to watch. To actively reach the device, just call notify/ask/confirm — they connect on demand and fail closed.

When the device is unavailable

All three tools return "unavailable: <reason>" when the Cardputer is off or out of BLE range (and a cloud call also fails if the bridge daemon or tunnel is down). For notify/ask, degrade silently to normal chat behavior. For confirm, the unavailable case is a hard stop (see §1) — never treat an unreachable safety device, a sleeping laptop, or a dead tunnel as implicit approval. Fail closed, every time.

What this skill is not

This is runtime etiquette only. Flashing firmware, pushing apps, and provisioning the device are handled by the separate m5-onboard skill — don't duplicate that here.

© dakshaymehta, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .claude/skills/cardputer-companion of dakshaymehta/cardputer-claude-os.

Open the folder on GitHubat commit 8e2bc8a

Compare with similar skills

Cardputer Companion next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Cardputer Companion compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Cardputer Companion this skilldakshaymehta/cardputer-claude-os208—~2.5kAutomated safety check: PassApache-2.0
MCP Server Builderanthropics/skills180k63 repos~2.3kAutomated safety check: PassApache-2.0
MCP Server BuildershareAI-lab/learn-claude-code78k4 repos~1.2kAutomated safety check: PassMIT
MCP Integration for Pluginsanthropics/claude-plugins-official38k11 repos~3.1kAutomated safety check: PassApache-2.0
Crush Configurationcharmbracelet/crush29k—~3.7kAutomated safety check: PassCustom licence
Context Mode Output Sandboxmksglu/context-mode26k—~4.1kAutomated safety check: PassCustom licence

Similar skills

  • MCP Server Builder

    anthropics/skills

    Official

    Guides the design and implementation of Model Context Protocol servers in TypeScript or Python, from tool naming and error messages to evaluation.

    180k GitHub starsUsed in 63 repos~2.3k tokens
    Agent WorkflowsAuto-check passed
  • MCP Server Builder

    shareAI-lab/learn-claude-code

    Walks through building MCP servers in Python or TypeScript that expose tools, resources and prompts to Claude, with templates, registration and testing.

    78k GitHub starsUsed in 4 repos~1.2k tokens
    Agent WorkflowsAuto-check passed
  • MCP Integration for Plugins

    anthropics/claude-plugins-official

    Official

    Explains how to bundle Model Context Protocol servers in a Claude Code plugin, covering config files, stdio, SSE, HTTP and WebSocket server types, and authentication.

    38k GitHub starsUsed in 11 repos~3.1k tokens
    Agent WorkflowsAuto-check passed
  • Crush Configuration

    charmbracelet/crush

    Explains how to configure the Crush coding agent with crushrc or crush.json, covering providers, models, LSPs, MCP servers, hooks, permissions and config precedence.

    29k GitHub stars~3.7k tokensUpdated today
    Agent WorkflowsAuto-check passed
  • Context Mode Output Sandbox

    mksglu/context-mode

    Routes large command, file, API and browser output through context-mode tools so only the needed result enters the agent's context, instead of dumping it via Bash.

    26k GitHub stars~4.1k tokensUpdated today
    Agent WorkflowsAuto-check passed
  • Migrates the compatible subset of settings and global file-based MCP servers from the Warp desktop app into Warp Agent CLI without exposing credentials or state.

    65k GitHub starsUsed in 1 repo~2.1k tokens
    Agent WorkflowsAuto-check passed

Categories

Questions about Cardputer Companion

What does Cardputer Companion do?

A skill your agent uses when the cardputer MCP tools (notify, ask, confirm, show, progress) are available in the session, or the user mentions their Cardputer / handheld / "buzz me" / "page me". Cardputer Companion is an agent skill from dakshaymehta/cardputer-claude-os. Use when the cardputer MCP tools (notify, ask, confirm, show, progress) are available in the session, or the user mentions their Cardputer / handheld / "buzz me" / "page me".

When should I use Cardputer Companion?

Cardputer Companion fits situations like: the cardputer MCP tools (notify; progress) are available in the session; the user mentions their Cardputer / handheld / buzz me / page me; even when the user doesnt name the skill — if the cardputer MCP tools are registered.

How do I install Cardputer Companion in Claude Code?

Run `npx skills add dakshaymehta/cardputer-claude-os --skill cardputer-companion -a claude-code`. Or copy the skill folder (.claude/skills/cardputer-companion in dakshaymehta/cardputer-claude-os) into .claude/skills/cardputer-companion in your project. Claude Code loads it when a task matches its description.

How do I install Cardputer Companion in Codex?

Run `npx skills add dakshaymehta/cardputer-claude-os --skill cardputer-companion -a codex`. Or copy the skill folder (.claude/skills/cardputer-companion in dakshaymehta/cardputer-claude-os) into .agents/skills/cardputer-companion in your project. Codex loads it when a task matches its description.

Can I use Cardputer Companion in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add dakshaymehta/cardputer-claude-os --skill cardputer-companion -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/cardputer-companion, .gemini/skills/cardputer-companion, .github/skills/cardputer-companion and .opencode/skills/cardputer-companion in your project.

What does Cardputer Companion need to run?

Going by SKILL.md and its folder, Cardputer Companion needs the command-line tools its instructions call (git and pytest).

Does Cardputer Companion access the network?

SKILL.md contains no URLs. Its commands use git, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is Cardputer Companion safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Cardputer Companion use?

Cardputer Companion is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Cardputer Companion use?

About 2.5k tokens (SKILL.md is roughly 10k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Cardputer Companion?

Skills that share tags, products or a category with Cardputer Companion: MCP Server Builder (anthropics/skills, 180k stars), MCP Server Builder (shareAI-lab/learn-claude-code, 78k stars), MCP Integration for Plugins (anthropics/claude-plugins-official, 38k stars) and Crush Configuration (charmbracelet/crush, 29k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Cardputer Companion?

dakshaymehta (a GitHub user) maintains it in dakshaymehta/cardputer-claude-os, which has 208 GitHub stars. The repository was last updated on July 1, 2026.

Source: dakshaymehta/cardputer-claude-os on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.