← back to work
case study · /slack-bots

Slack bots for the QA stack.

Two ChatGPT Enterprise bots built on connectors / MCP / skills. One for triage. One for generating test cases. Both deployed cross-team so PMs and devs can use QA tooling without leaving Slack.

production · lotusflare chatgpt enterprise slack connectors · MCP · skills
↗ medium ↗ linkedin

"Triage this for me."

Drop a Jira ticket link or paste a stack trace into the channel and tag the bot. It does what the Multimodal Triage pipeline does, but conversational:

  • Fetches the ticket via the Jira connector.
  • If there's a video attachment, runs the Gemini 3 extraction.
  • Cross-references Sentry + Victoria Logs via MCP.
  • Surfaces likely commits via GitHub MCP.
  • Returns a Slack-formatted summary with confidence.

What it doesn't do: act. The triage bot only summarizes. Humans decide.

"Write me a test case for X."

The AIO Test Case Creator skill, but in Slack. PMs and devs can:

  • Describe a flow in plain language ("user activates eSIM with number transfer on a family line").
  • Get back a fully-normalized test case in the canonical format.
  • One-click "publish to AIO" once it looks right.

The lift: PMs and devs no longer wait for QA to write tests for new features. They draft. QA reviews. Releases speed up.

The guard rail: every test created via the bot lands in Status: Draft and goes through standard human review before promotion.

Why ChatGPT Enterprise vs. a custom bot

  • Connector-native. Jira, GitHub, Slack, Confluence all wire in without extra OAuth juggling.
  • Identity-scoped. Each user's queries inherit their own access. A PM can't accidentally see a sensitive Jira project they don't have access to.
  • MCP and skills are first-class. The same skill that powers AIO TCC plugs into the bot directly — no duplication.

Time to first useful bot: about a week. Without ChatGPT Enterprise as the substrate, that would've been months.