keypick

name

keypick

description

Use this skill whenever the user needs API keys, secrets, or environment variables to run code, scripts, or shell commands — even if they don't mention KeyPick by name. Trigger on phrases like "run this script", "test the API", "it needs my key", "set up my env", "what keys do I have", or any task that requires injecting credentials. Also trigger for vault management: listing stored keys, adding new ones, extracting to .env, or switching vaults. If the user has KeyPick installed and needs any kind of secret, this skill applies.

KeyPick — Claude Code Integration

KeyPick is a biometric-secured API key vault. Use it to inject keys into commands, discover what's stored, and manage the vault — without ever asking the user to paste a secret into the chat.

Quick command reference

Goal	Command	Auth required?
Inject keys into current shell	eval $(keypick auto <Group>)	No
List all groups and key names	keypick list	Yes (biometric)
Add or update keys	keypick add	Yes (biometric)
Copy one key to clipboard	keypick copy	Yes (biometric)
Extract groups to a .env file	keypick extract	Yes (biometric)
Show/switch active vault	keypick vault current / keypick vault select	No
Manage per-project .env in vault	keypick env status/push/pull	No / Yes
First-time setup	keypick setup	No

Injecting keys before a command

keypick auto <Group> [Group2 ...] is the non-interactive path — no biometric prompt, designed for scripting. It prints export KEY='VALUE' lines to stdout.

# Inject one group, then run the command
eval $(keypick auto OpenAI) && python script.py

# Inject multiple groups at once
eval $(keypick auto Stripe OpenAI) && node server.js

# Verify a key is present WITHOUT revealing its value
eval $(keypick auto Stripe) && echo "STRIPE_API_KEY=${STRIPE_API_KEY:+set}"

The ${VAR:+set} form prints the literal word set only when the variable is populated — it never echoes the secret itself. Never use env, printenv, set, or env | grep to verify injection: those print the plaintext values into your tool output, which means the secret enters the conversation context.

When you don't know the group name: ask the user — "Which KeyPick group holds those keys?" — or suggest they run keypick list themselves (it requires their fingerprint so you can't run it silently).

When the group name is obvious from context (e.g., the script imports openai, the user mentions "Stripe"), make a reasonable inference and tell the user what you're using: "I'll pull keys from your OpenAI group — let me know if that's named differently in your vault."

Running a command that needs keys

The standard pattern for any Bash tool invocation that requires secrets:

eval $(keypick auto <Group>) && <the actual command>

Prefer this over a multi-step approach — it keeps keys scoped to that single subshell invocation and they don't linger in the environment.

Vault management operations

These require the user's biometric — run them interactively and let the user authenticate when prompted.

keypick list          # Shows all groups and key names (values hidden)
keypick add           # Prompts to add/update a group
keypick copy          # Interactive: pick group → pick key → copies to clipboard
keypick extract       # Interactive: pick groups → writes a .env file
keypick vault select  # Choose a different vault repository

For keypick env push/pull, the user must be in the project directory that owns the .env files.

After keypick extract, a plaintext .env exists on disk. Do not cat, Read, or Grep that file from the AI side — see security notes below. If the user only needs a single value rather than a whole group, suggest keypick copy instead.

Checking if KeyPick is installed

keypick --version 2>/dev/null || echo "not installed"

If not installed, direct the user to run the installer:

• Windows: irm https://raw.githubusercontent.com/seanrobertwright/KeyPick/master/install.ps1 | iex
• macOS/Linux: bun install -g keypick

If installed but not set up: keypick setup

Security notes — keeping secrets out of AI context

The eval $(keypick auto …) && <command> idiom is safe by construction: $(…) captures the export lines so they never reach your tool's stdout, and the subshell dies when the command finishes. That guarantee is fragile — it only holds if you don't do any of the following.

Hard rules — these commands leak plaintext into the conversation:

• Do not run env, printenv, set, env | grep, or any sibling command whose purpose is to display environment variables. Their stdout is captured and returned to the model.
• Do not cat, Read, Grep, or otherwise ingest .env, .envrc, .env.local, or any file that may contain plaintext secrets — including files produced by keypick extract. Treat those files as write-only from the AI's perspective: created for the user's process, never read back.
• Do not run commands that print env vars as part of their own output (e.g., node -e 'console.log(process.env)', debug middleware that dumps config, verbose CI loggers). If a script fails and the error includes env dumps, stop and tell the user rather than pasting the output back.
• Do not use set -x, bash -x, or other shell trace modes while secrets are in scope — trace output expands eval and prints the values.

Preferred patterns:

• Prefer eval $(keypick auto …) && <command> over export-ing into a persistent shell session — the subshell scope is the isolation.
• When the user only needs one value for their own use (pasting into a dashboard, a config UI, etc.), suggest keypick copy instead of keypick extract. copy puts the value on the OS clipboard and never writes plaintext to disk.
• When you must verify injection worked, use echo "KEY=${KEY:+set}" — prints the literal word set without revealing the value.

What KeyPick itself guarantees:

• keypick auto never writes to disk; keys live only in the calling subshell.
• keypick list shows key names only — values are hidden.
• keypick copy writes the value to the OS clipboard; only the group/key name is printed.
• Biometric-gated commands (add, extract, copy, list) cannot be run non-interactively by the AI — the user has to approve at the OS prompt.