Approve or deny from the lock screen, from any script
For Codex, Cursor, n8n and your own scripts, which ask through Monish’s API. Using Claude Code? Its permission prompts reach your iPhone through a hook instead: see Approve Claude Code permissions from your iPhone.
Quick answer
Your agent sends Monish a question, a message of type: ask. It shows on your lock screen as a live card with Approve and Deny, and as a notification. You tap; your agent, which was waiting for the answer, reads approved or denied and carries on or stops. No answer before the deadline counts as a no.
What a question carries
| Field | What it does |
|---|---|
title | The question itself, up to 200 characters. Keep it short: the lock screen shows about one line until Monish 1.0.1. |
body | The stakes in a line, up to 200 characters: “3 commits · tests green”. Never the question again. |
fields.context | One more line of context, up to 120 characters. |
fields.risk | low (easy to undo), medium (reversible but visible or costly) or high (cannot be undone: real people, money, deletion). The app shows it in the decision sheet, high in red. |
fields.approve_labelfields.deny_label | Rename the two buttons, up to 24 characters each: “Ship it” and “Not today”. Approve and Deny by default. |
expires_in | Seconds before the question expires, from 300 to 86,400. One hour by default. |
id | Your handle on the question: the agent reads the answer at /v1/answer/<id>. |
agent | The name shown on the card, such as Codex · lumen. |
The exact JSON
One request to send the question. The key goes in the authorization header, or as "key" in the body.
Send the question
curl -s https://monish.cc/v1/send \
-H 'content-type: application/json' \
-H "authorization: Bearer $MONISH_KEY" \
-d '{
"id": "deploy-lumen-2-4",
"type": "ask",
"agent": "Codex · lumen",
"title": "Ship lumen 2.4 to production?",
"body": "312/312 tests green · 2 migrations",
"expires_in": 900,
"fields": {
"context": "3 commits since 2.3 · rollback ready",
"risk": "medium",
"approve_label": "Ship it",
"deny_label": "Not today"
}
}'
The server answers 202
{"id":"deploy-lumen-2-4","status":"queued"}
Then the agent waits. Each call to /v1/answer holds for up to 25 seconds and returns pending if nothing came, so the loop is gentle, and it ends at the latest when the question expires.
Wait for the answer
while :; do
s=$(curl -s https://monish.cc/v1/answer/deploy-lumen-2-4 \
-H "authorization: Bearer $MONISH_KEY" | jq -r .status)
[ "$s" != pending ] && break
done
echo "$s" # approved, denied, expired, or text (then read .text)
What the agent reads once you tap
{"id":"deploy-lumen-2-4","status":"approved","answered_at":"2026-10-27T15:42:10Z","text":null}
What happens when you tap
- You tap Approve or Deny on the lock-screen card or on the notification, without opening the app.
- Monish records the first answer. A second tap, on another surface, changes nothing.
- The live card closes, and your agent’s next check returns
approvedordenied. - Rather answer in words? Open the question in the app and type a reply, up to 500 characters. The agent gets
textand reads it as an instruction: “yes, but 30 %”, “after the release”.
If you don’t answer
When expires_in runs out, the question expires: the card on your lock screen ends as expired (shown as failed before 1.0.1), without its buttons, and the agent reads expired. The skill tells agents to treat it as a no: stop that step, say so, and never ask the same thing again. The server never lets a question expire in under 5 minutes, and keeps answers 24 hours before deleting them.
Examples
Ask for decisions you would want to make yourself, not for “shall I continue?”.
In code
- Deploy harbor 1.6 to production?ApproveDeny
Deploy · risk medium: reversible, but visible.
- Open a PR for 4 commits?Open PRHold
Pull request · risk low: nothing lands until you merge.
- Run the full eval on 2,000 prompts, about $38?Run itSkip
Spend · risk medium: costs money, bounded.
In life
- Buy the two train tickets to Lyon, €84?BuyNot yet
Purchase · risk high: real money.
- Send this reply to your landlord?SendEdit first
Email · risk high: a real person reads it.
- Book the 8:30 pm table for 4 at Chez Léon?BookSkip
Booking · risk low: free to cancel.
To get there, tell your agent once: “Before you deploy, send or spend, ping me first.”
Questions
- Do I need to open the app to answer?
- No. Approve and Deny work from the lock-screen card and from the notification. Open the app only when you want to type a reply.
- Can my agent do anything else on my phone?
- No. Your agent can send items to Monish and read the answer to its own questions. It cannot read or control anything else on your iPhone.
- What if two questions arrive at once?
- Each has its own id and its own card. Questions stay at the top of the feed until you answer them or they expire.
- Which agents can ask?
- Anything that can run curl: Codex, Cursor, an n8n workflow, a cron job or a shell script. Claude Code on a paired computer asks through its hooks instead.
More guides
- Claude Code notifications on your iPhone: setupPair in one command, then allow, deny or reply from the lock screen.
- Codex notifications on your iPhoneOne rule in AGENTS.md, and a hook that rings when Codex waits.
- n8n notifications on your iPhoneOne HTTP Request node: end of a workflow, errors, approval before an email.
- How Monish comparesPrice, account, Live Activities and data kept, side by side.
New to Monish? The home page shows what lands on your iPhone, and all guides are listed here.
Get Monish on your iPhone
Free, no account, no SDK. Your key is waiting in the app, inside the line to paste.