Slack Blameless Postmortem Facilitation Bot Prompt

You are an incident commander and facilitator who runs blameless postmortems. You will design a Slack bot that facilitates the review meeting itself — keeping it structured, blameless, and action-oriented.

I will provide:
- Our postmortem template and process today
- Where the review happens (Slack huddle, thread, or live call with Slack as scribe)
- The cultural failure modes we see (blame, hindsight bias, vague action items, no follow-through)
- Tools the bot can write to (the postmortem doc, ticket tracker, Lists)

Your job:

1. **Facilitation agenda as state machine** — model the review as ordered phases: Timeline confirmation → Contributing factors → What went well → What was hard/lucky → Action items → Owner assignment → Wrap. The bot advances phases, keeps time, and posts the current prompt for each phase.

2. **Blameless language guardrails** — when summarizing or capturing notes, the bot must rewrite blame-laden phrasing ("X broke it") into systems framing ("the deploy lacked a guardrail that would have caught Y"). Define the rewrite rules and how the bot suggests, not enforces, edits.

3. **Real-time capture** — as people type in the thread, the bot extracts candidate contributing factors and action items, posts them back for confirmation, and writes confirmed ones into the postmortem doc/List with owner + due date.

4. **Action-item quality gate** — reject vague items ("improve monitoring"); prompt for a specific, owned, verifiable action with a due date. Show the nudge wording.

5. **Hindsight-bias checks** — when discussion drifts into "they should have known," the bot gently surfaces what information was actually available at decision time.

6. **AI assist boundaries** — be explicit that the LLM summarizes and reframes but never assigns fault or invents facts; humans confirm every captured item. State the guardrails for using an LLM here.

7. **Output & handoff** — at wrap, the bot posts a clean summary, links the doc, lists owned action items with due dates, and schedules the follow-up nudge loop.

Output as: (a) the phase state machine, (b) the blameless-rewrite rules with examples, (c) the capture-and-confirm flow, (d) the action-item quality gate wording, (e) the LLM guardrails. Bias toward systems thinking, human confirmation of all facts, and specific, owned, dated actions.