1Password CLI

Follow the official CLI get-started steps. Don't guess install commands.

References

• references/get-started.md — install + app integration + sign-in flow
• references/cli-examples.md — real op command examples

Existing environment setup

The user's shell profile already handles several things before Claude starts. Do not duplicate this work — build on it instead.

Pre-injected at launch (via cc / cc-dsp shell wrappers in ~/.zshrc):

• GITHUB_PERSONAL_ACCESS_TOKEN is exported via op read before Claude starts. It's already available in the environment — don't re-read it.
• STITCH_API_KEY is exported via op read before Claude starts. Used by the Stitch AI MCP server — don't re-read it.

SSH agent:

• SSH_AUTH_SOCK points to the 1Password SSH agent socket. SSH operations (git clone/push over SSH, etc.) work automatically.

Default vault: agentic_ai

• All secret references should use op://agentic_ai/<Item Name>/<field> unless the user specifies a different vault.
• This convention is also documented in ~/.claude/CLAUDE.md.

Adding secrets for MCP servers

Same pattern as GITHUB_PERSONAL_ACCESS_TOKEN:

1. Store in 1Password: op item create --vault="agentic_ai" ...
2. Add export VAR="$(op read 'op://agentic_ai/...')" to cc/cc-dsp in ~/.zshrc
3. Reference as ${VAR} in the .mcp.json env block

That's it — no wrapper scripts needed. The env var is injected before Claude starts, and .mcp.json expands ${VAR} references in env values.

When you need this skill

Use this skill for mid-session secret operations:

• Reading a secret not already in the environment (e.g., a database password, API key for a new service)
• Storing a new secret the user just generated
• Injecting secrets into a config template (op inject)
• Wrapping a command with secrets (op run)
• Troubleshooting op auth or connectivity issues

Workflow

1. Check if the secret is already in the environment (printenv VAR_NAME). If it's there, use it — no need for op.
2. Verify CLI is present: op --version.
3. Try op whoami to check auth status. If desktop app integration is enabled and the app is unlocked, this may just work without tmux.
4. If op whoami fails or returns "not signed in", set up a tmux session (see below) and authenticate there.
5. For multiple accounts: use --account or OP_ACCOUNT.

tmux session for persistent auth

Claude Code's shell tool spawns a fresh TTY per command, which can lose op authentication state between calls. A persistent tmux session solves this by keeping a single authenticated shell alive.

Use tmux when:

• op whoami fails outside tmux
• You need to run multiple op commands in sequence
• Sign-in requires interactive authorization (app prompt)

Example setup:

SOCKET_DIR="${TMPDIR:-/tmp}/claude-op-sockets"
mkdir -p "$SOCKET_DIR"
SOCKET="$SOCKET_DIR/op-session.sock"
SESSION="op-auth-$(date +%Y%m%d-%H%M%S)"

tmux -S "$SOCKET" new -d -s "$SESSION" -n shell

# Authenticate
tmux -S "$SOCKET" send-keys -t "$SESSION":0.0 -- "op signin" Enter

# Verify (wait a moment for the app prompt to complete)
sleep 2
tmux -S "$SOCKET" send-keys -t "$SESSION":0.0 -- "op whoami" Enter
sleep 1
tmux -S "$SOCKET" capture-pane -p -J -t "$SESSION":0.0 -S -50

# Run your op commands inside the session
tmux -S "$SOCKET" send-keys -t "$SESSION":0.0 -- "op read 'op://agentic_ai/MyItem/password'" Enter
sleep 1
tmux -S "$SOCKET" capture-pane -p -J -t "$SESSION":0.0 -S -10

# Clean up when done
tmux -S "$SOCKET" kill-session -t "$SESSION"

Common operations

Read a secret

op read 'op://agentic_ai/<Item Name>/<field>'

Export as environment variable

export VAR_NAME="$(op read 'op://agentic_ai/<Item Name>/<field>')"

Store a new secret

Prompt the user — never write the value to disk:

op item create --category=api-credential --title="<Service Name>" --vault="agentic_ai" '<field>=<value>'

Inject into a config template

op inject -i config.yml.tpl -o config.yml

Wrap a command with secrets

op run --env-file="./.env" -- <command>

Note: op run doesn't allocate a TTY. For interactive tools, prefer op read + export instead.

Guardrails

• Never paste secrets into logs, chat, or code.
• Never write secrets to disk — prefer op run / op inject / op read + export.
• If you discover plaintext secrets in any file, flag it immediately and offer to migrate them to the agentic_ai vault.
• If a command returns "account is not signed in", authenticate inside tmux and re-try.
• If tmux is unavailable, ask the user before proceeding — don't attempt op commands that will silently fail.