Tutorial

How to Start Coding Agents From Slack for Bug Reports and Small Fixes (2026)

Turn a Slack bug report or request into a Claude Code or Codex run that reproduces the bug and opens a draft pull request. Paradev's Slack bot and Slack triggers, Claude Code in Slack, Codex, and Cursor compared.

Download Paradev

For macOS on Apple Silicon and Intel. Linux AppImage is experimental.

Last updated·

Paradev was called Superset until October 2026.

To start a coding agent from Slack, connect Slack to a tool that turns a message into an agent run, and have the agent post its result back as a draft pull request or a short note. Paradev gives a team two ways to do this. Mention the Paradev Slack bot with a request, and it can file a task or start a coding agent and report back in the thread. Or set an Automation (Pro) with a Slack trigger, such as every new message in #bugs or any message someone reacts to with :bug:, and Paradev starts Claude Code, Codex, or another CLI agent on a machine you choose.

The vendor options are Claude Code in Slack, which turns an @Claude coding request into a Claude Code cloud session; ChatGPT in Slack, where an @ChatGPT mention sends repository work to a Codex Cloud task; and Cursor in Slack, where an @Cursor mention starts a Cursor Cloud Agent. Each of those runs one vendor's agent, in that vendor's environment by default. Paradev runs the agent you choose on your own machines, on a Linux server in your own cloud account with Bring Your Own Cloud, or in a cloud workspace.

This guide is for a team whose bug reports and small requests already arrive in Slack. It is published by Paradev. Details about other tools come from their own docs as of October 2026.


Which Slack messages are worth handing to an agent?

Good candidates have a clear expected result and a small surface: "the export button returns 500 on reports over 10k rows", "bump the Sentry SDK to v9 across all apps", "the settings page shows the wrong timezone". Poor candidates need a product decision or a long discussion first. A simple team rule helps: a person decides which messages go to an agent, with a mention or an emoji, and the agent never merges.

Option 1: mention the Paradev Slack bot

Connect Slack in the dashboard under Integrations → Slack (OAuth). For code changes, the bot also needs what any cloud workspace needs: GitHub connected in Integrations, with the Paradev GitHub App given access to the repository, and your agents signed in under Settings → Agents. Then mention the bot in a channel or send it a DM:

For a code change, the bot starts a cloud workspace, launches a coding agent there, and reports back in the thread. For a request like "file a bug and assign it to whoever owns reporting", it creates the task instead, with assignee, status, and priority from your Linear or Paradev setup. It reads the thread and the channel history on its own, so you do not paste earlier messages into the request.

Once the bot has replied in a thread, and thread follow-ups are on for your Slack team, replies in that thread reach the bot without another mention. !stop ends the bot's current Slack turn, but it does not stop a coding agent the bot already started in a workspace; open that workspace to stop it. !mute brings back mention-only mode for that thread.

Two limits: code changes from the bot run in cloud workspaces, which come with the Team and Ultra plans. And the bot's own tools for Linear and GitHub cannot delete resources, merge pull requests, or make bulk writes.

Option 2: a Slack trigger that runs your agent on your machine

An automation with a Slack trigger runs on the device you pick, with the agent you pick, against your repository.

  1. Connect Slack under Integrations.
  2. Go to Automations → New automation. Pick the project, a device that stays online, and an agent such as Claude Code or Codex.
  3. Under Triggers, add Slack → Message in channel and pick #bugs. Optionally set the message filter to a word your team puts in reports, like bug. Or add Reaction added with an emoji like :bug:, so people choose which messages the agent picks up. Press Save. The bot joins public channels when you save; for a private channel, invite it first.
  4. Write a prompt that works from the report and stops when it should:

From the CLI, find the channel ID and create the same automation with a reaction trigger:

Without --host, the automation runs on the machine where you type the command. Pass the ID of a host that stays online, from superset hosts list.

What to know about Slack triggers:

  • Top-level messages only. Thread replies do not fire Message in channel, so a discussion under a report does not start more runs.
  • No loops from the bot. Slack triggers ignore the Paradev bot's own messages and reactions.
  • The device must be online. A run that fires while the machine is asleep is recorded as missed and is not retried.
  • Replying in Slack. The run result lives in a Paradev workspace. To have the agent answer in the thread itself, give it a Slack tool, such as the Slack plugin.

The Slack bug report recipe has more variations.

Which option should a team pick?

Paradev Slack botParadev Slack triggerClaude Code in SlackChatGPT (Codex) in SlackCursor in Slack
Started byA mention or DMA message or emoji in channels you pickAn @Claude mention in a channelAn @ChatGPT mention in a channelAn @Cursor mention in a channel, or a DM
AgentA coding agent in a cloud workspaceAny CLI agent you set, such as Claude Code or CodexClaude CodeCodexCursor Cloud Agent
Runs onCloud workspaceYour machine, a shared host, or a cloud workspaceClaude Code cloud sessionCodex Cloud taskCursor's managed infrastructure by default; Team Pools on Enterprise
ResultReply in the threadA workspace, plus a draft PR or note if the prompt asksStatus updates in the thread, then View Session and Create PR buttonsThe result in the thread, with a Follow along link to the taskStatus updates in the thread, then a notification with the PR it created
Good fitAd hoc requests and task filingA steady stream of reports in one channelPro, Max, and Ultra accounts in a Slack workspace not connected to Claude TagChatGPT workspaces with Codex Cloud environments set upTeams already on Cursor

Per Anthropic's Claude Code in Slack docs, it works in channels, not DMs, needs cloud sessions and a connected GitHub account, and picks the repository from the conversation. This version answers Pro, Max, and Ultra accounts; Team and Enterprise plans use Claude Tag instead. Anthropic also notes that Claude may follow directions from other messages in the context, so use it only in trusted conversations. Per OpenAI's ChatGPT in Slack docs, @ChatGPT replaces the earlier @Codex mention, and repository work needs cloud delegation and a published environment shared with the workspace. Per Cursor's Slack docs, it needs a connected repository provider, a default repository, and usage-based pricing. The same caution applies to any agent that reads Slack: a Slack message is untrusted input, so keep runs on draft pull requests and keep merge rights with people.

Frequently Asked Questions

Can Claude Code run from a Slack message?

Yes. Claude Code in Slack starts a Claude Code cloud session from an @Claude mention, for Pro, Max, and Ultra accounts. To run Claude Code on your own machine instead, use a Paradev automation with a Slack trigger and pick Claude Code as its agent.

Can a Slack emoji start a coding agent?

Yes. In Paradev, a Reaction added trigger with an emoji such as :bug: starts a run for any message that gets that reaction in the channels you pick.

Does the agent reply in the Slack thread?

The Paradev bot does. An automation's run lives in a Paradev workspace; give the agent a Slack tool if you want it to post in the thread too.

Can I use Codex instead of Claude Code?

Yes. Set Codex as the automation's agent, or pass --agent codex on the CLI.

Which plan do I need?

The Slack integration and automations are on Pro. Cloud workspaces come with the Team and Ultra plans. See pricing.