AGENT SKILL

comms-update

Draft an incident communication update — plain language, status (investigating/identified/monitoring/resolved), impact, next-update ETA.

License 许可
MIT
最近更新
2026-10-29
Stars
★ 0
Forks
0

01 前言 Front Matter

name
comms-update
description
Draft an incident communication update — plain language, status (investigating/identified/monitoring/resolved), impact, next-update ETA.

03 SKILL.md 正文

comms-update

Draft one comms update (status page / customer email / internal stakeholder note). Output is the update, ready for the comms lead to approve — this skill drafts; posting is a separate, approved step (polymath-chat:post-incident-comms, polymath-chat:post-statuspage-update).

Mode: analytical — judgment inside a fixed frame. Keep the output shape; every verdict cites its evidence; unknowns are marked [TBD] or [VERIFY], never filled in.

When to use

  • A sev1/sev2 needs an external update.
  • Stakeholders need an internal update during a long incident.
  • A workflow's comms step generates a draft (e.g. respondToIncident paired with polymath-chat:post-incident-comms).

Inputs

  • The incident's current facts (required): what users see, start time, scope, mitigation status — from the incident channel or the IC. Anything unconfirmed is left out or stated as unknown, never guessed.
  • Audience (required): external (status page, customers) or internal stakeholders.
  • The previous update (optional) — the new one says what changed since.
  • Drafting is not posting: publishing goes through polymath-chat skills, which confirm before sending.

Procedure

  1. Status — pick one from a fixed scale:
    • investigating — we know there's a problem; we don't know the cause.
    • identified — we know the cause; mitigation in progress.
    • monitoring — we believe we've mitigated; watching to confirm.
    • resolved — service is healthy; postmortem to follow.
  2. Audience — externals (status page, customer email) get different language than internal stakeholders. Pick once.
  3. Plain language — what users see and can/can't do. Not "ELB target group degraded".
  4. Workarounds — if any, name them simply.
  5. Next-update ETA — always specify. "Next update in 30 minutes" beats silence.
  6. What we know / don't know — both sections present. Don't speculate.

Output

Update #3 — Refund processing degraded

Status: identified

What's happening:
  Customers attempting to issue a refund are seeing errors. New refunds
  are being queued internally and will process once we restore service.
  Existing orders, account access, and other operations are unaffected.

Impact:
  - All customers attempting refunds since 14:02 UTC.
  - Refunds queued during this window are expected to process after
    recovery; we will confirm once the queue has drained.

What we're doing:
  Rolled back the recent refund-service deploy. Metrics improving; we
  are watching to confirm recovery.

What we know:
  A bad deploy of refund-service introduced a regression in the
  payment-provider integration. The rollback restores prior behavior.

What we don't know yet:
  Whether the queued refunds will all process within the next 15
  minutes (estimated based on queue depth).

Next update in 15 minutes, at 14:35 UTC.

Quality bar

  • One of the four fixed status values.
  • Next-update ETA stated.
  • "What we don't know yet" present and honest.
  • No internal-jargon component names.
  • No "due to a database issue" — that's the cause, not the impact.

Anti-patterns to avoid

  • "We are aware of an issue" without saying what's broken.
  • Silence between updates.
  • Speculation about root cause in the comms update.
  • Promising a fix ETA you can't keep.

04 同仓库更多 Skills