Skip to content
Back to skills

Render Cli

ASecurity

Use the Render.com CLI v2.x (`render`) to inspect Render services, check CLI auth, list services and deploys, trigger deploys when explicitly requested, and view logs. Use when an agent needs Render CLI commands, non-interactive flags, deploy status polling, or service auto-deploy checks.

  • 71 stars
  • 0 votes
  • 0 copies
  • 8 views
  • Added September 3, 2026
ai-agentsgobashgit

Works with

  • terminal
  • cli

Security analysis

A100/100

Scanned September 3, 2026

npx -y skills add atrislabs/atris --skill render-cli --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Render Cli?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Render Cli
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/atrislabs-render-cli/badge)](https://www.skillsdirectory.com/skills/atrislabs-render-cli)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: render-cli
description: Use the Render.com CLI v2.x (`render`) to inspect Render services, check CLI auth, list services and deploys, trigger deploys when explicitly requested, and view logs. Use when an agent needs Render CLI commands, non-interactive flags, deploy status polling, or service auto-deploy checks.
version: 1.0.0
tags:
  - render
  - deploy
  - cli
---

# Render CLI

Use the `render` binary to manage Render services, deploys, logs, and sessions from the terminal. Treat it like a production-control CLI: read first, parse JSON, and only create or mutate resources when the operator explicitly asked for that action.

## Agent defaults

Always add these flags for agent-safe commands:

- `--output json` for parseable output.
- `--confirm` to skip confirmation prompts.

The CLI defaults to interactive mode and can hang an agent if a command prompts. If a command is intended to run unattended, include both flags unless the CLI help says the mode is interactive-only.

## Auth

Check the active login before assuming access:

```bash
render whoami --output json --confirm
```

If the user is not logged in, ask them to complete:

```bash
render login
```

`render login` uses the dashboard/browser login flow, so do not run it in an unattended automation loop.

## Services

List services in the active workspace:

```bash
render services --output json --confirm
```

Use the returned service `id` as `<service-id>` in deploy and log commands. When a push to Git did not deploy, inspect the service JSON for `autoDeploy`; it may be disabled for that service.

## Deploys

Trigger a deploy only when the operator explicitly asks for a real deploy:

```bash
render deploys create <service-id> --output json --confirm
```

Watch deploy state by polling:

```bash
render deploys list <service-id> --output json --confirm
```

Common deploy statuses include `queued`, `build_in_progress`, `update_in_progress`, `live`, and `failed`. Treat `live` and `failed` as terminal states for ordinary deploy monitoring.

## Logs

Query recent logs for a service:

```bash
render logs --resources <service-id> --limit 100 --output json --confirm
```

For a live tail, use:

```bash
render logs --resources <service-id> --tail --confirm
```

The CLI help says `--tail` streams only in interactive mode, so prefer bounded JSON log queries for automated agents and use live tail only when a long-running interactive stream is acceptable.

## Gotchas

- The CLI is interactive by default; `--confirm` prevents confirmation prompts from blocking the agent.
- `--output json` makes service, deploy, and log output machine-readable.
- In non-interactive log mode, `--resources <service-id>` is required.
- Git pushes do not guarantee deploys. Check `autoDeploy` in the service JSON before assuming push-to-git should deploy.
- Do not use placeholder IDs in real commands. Resolve a service ID from `render services --output json --confirm` first.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.