slack

Slack Tools

Overview

Tools for working in Slack. The investigation tools (list_slack_channels, list_slack_users, get_channel_history, get_thread_replies, get_connected_slack_channels) are read-only and used during postmortem generation and interactive investigation to gather human context (deployment decisions, communication gaps, resolution steps). In Agent mode you also get post_slack_message to post a message or threaded reply.

The tool signatures and parameters are provided to you in the tool schema — this skill only covers when to use each and Slack-specific behaviour.

Choosing a tool

• get_connected_slack_channels is the routing-decision source: the channels Aurora is aware of, each with a description of what it's for and which team/service it serves. Use it to pick which channel(s) are relevant when posting about an incident or notifying a team — and combine it with the Slack behaviour memory (below), which holds the team/channel routing preferences the descriptions alone don't capture.
• list_slack_users finds people to notify. Pass a name prefix (a for Ada and Alex). Copy the returned <@USERID> into the message. A plain @name does not notify, and an id you invent will not either.
• list_slack_channels is a live, description-less listing of channels the bot can access — use it for discovery (scan names/topics for a service or "incident"/"oncall"/"alerts"), not routing.
• get_channel_history / get_thread_replies read messages; scope history to the incident time window and follow into a thread when reply_count > 0.
• post_slack_message is the one write tool. Post a new message, or set thread_ts to reply UNDER an existing message. Available in Agent mode only.

Posting like a teammate

When you post about an incident, behave like a human on-call would — don't just dump a card into every channel:

1. Decide who cares using get_connected_slack_channels + the Slack memory. If no channel is relevant, stay silent (post nothing).
2. Before posting, read the recent history of the target channel (get_channel_history) and check whether this incident is already being discussed — your own earlier message, an incident.io/PagerDuty thread, or a human asking about it.
3. Recurring incident (you've seen it before — check the Incident Index in your prompt and your prior messages): reply IN THE THREAD of the existing message with a short note (e.g. "Still happening — 3rd time today, same DB pool exhaustion") using thread_ts. Do not start a new top-level message.
4. New incident: post a new, short message. Match the channel's preferred format from the Slack memory (some teams want a plain human line, others a short structured summary).
5. Keep it terse. One or two lines beats a wall of text.