Incident Comms

Incident Comms

When something is breaking or just broke, the comms are the product. This skill drafts the words a PM owns during an incident — grounded in the discipline behind Google's SRE incident response and Atlassian's incident handbook: a single comms owner, a predictable update cadence, say only what you know, and stay blameless. It produces four artifacts on demand — an internal escalation note, customer-facing status updates, a resolution/all-clear, and a leadership postmortem — not an engineering root-cause analysis.

Grounded in: Site Reliability Engineering — Beyer et al. (Google) + Atlassian's Incident Handbook: single incident commander, designated comms lead, fixed status cadence, blameless postmortem. Go deeper (The Product Channel): The Product Channel

> Not this skill: pre-mortem (imagine failure before you ship), release-notes (announce what shipped, on a good day), stakeholder-update (routine status when nothing is on fire). This skill is reactive — use it only when something is going wrong or just went wrong.

When to use this

• Production is degraded or down, a launch broke, or customers are hitting errors, and you need words out now.
• You need a customer-facing status-page or in-app notice that's honest without over-promising a fix time.
• Leadership is asking "what's going on?" mid-incident and you need one escalation note that aligns everyone.
• The incident is resolved and you owe customers an all-clear and leadership a blameless postmortem.
• A vendor/dependency outage is hitting your users and you have to communicate something you don't fully control.

Before you start (gather these)

• What's broken, for whom — the user-visible symptom (not the stack trace), and which segment/feature/region is affected.
• Severity & scope — how many users, is it total or partial, is data or money at risk. (SEV1 = broad/critical, SEV2 = significant, SEV3 = minor/contained.)
• Timeline so far — when it started, when you detected it, what's happened since (in UTC or one stated zone).
• Current status — investigating / identified / mitigating / monitoring / resolved, and the next concrete step.
• Owners — who is the incident commander (decides), who is the comms owner (you, usually), who is the customer audience.
• Constraints — is there a known ETA (rarely — don't invent one), a workaround, contractual/SLA or security/privacy implications.

If two or more of these are missing or vague, ASK 2-4 sharp questions before drafting — e.g., "What does the user actually see — error, slowness, wrong data, or nothing loading?", "Roughly what share of users, and is anyone's data at risk?", "When did it start and when did we detect it?", "Where are we now — still investigating, or do we have a fix deploying?" If you have enough to write something true, proceed and state your assumptions inline. Never invent an ETA, a root cause, or a customer count to fill a gap — "we don't yet know X" is a valid, trust-preserving line.

Process

1. Set severity and the comms cadence first. Severity drives how often you update and who you wake up. Lock a cadence and commit to it in the first message: SEV1 → every 30 min, SEV2 → every 60 min, SEV3 → at status changes. The next update time is a promise — meeting it, even with "no change," is what builds trust.
2. Name the single comms owner. One person speaks publicly (you). The incident commander runs the response; you translate it outward. Don't let five people post five versions.
3. Write the internal escalation note before anything customer-facing. Leadership and support need the full picture (scope, impact, who's on it, the ask) so they stop pinging the responders. Internal can be blunt; customer-facing must be careful.
4. Draft the customer update at the right altitude — say what you know. State impact in user terms, what you're doing, and when the next update lands. Acknowledge → impact → action → next-update-time. Omit internal hostnames, blame, speculation, and any ETA you can't stand behind.
5. Match honesty to the audience, not to your discomfort. Customers don't need the cause; they need to know you see it, you're on it, and when they'll hear from you again. Under-claim on the fix, over-deliver on the cadence.
6. Keep updating on cadence until resolved — even "still investigating" is an update. Silence reads as "they don't know it's broken" or "they've given up." A boring on-time update beats a detailed late one.
7. Send the resolution / all-clear when it's actually fixed — confirmed by monitoring, not hope. State what's restored, residual effects (e.g., delayed data catching up), any user action needed, and a brief, sincere apology proportional to the impact. Promise the postmortem if SEV1/2.
8. Write the blameless postmortem for PM/leadership, not eng. Lead with customer impact and the one-line cause in plain language. Frame the timeline and prevention as system/process fixes, never "Marcus deployed the bad config." Owners and dates on every action item, or it's theater. This is a leadership decision document, not the engineering RCA — point to that doc, don't reproduce it.

Output template

Pick the artifact(s) you need. Each numbered block is independently copy-pasteable. Fill in every [bracket]. Use one consistent timezone and label it.

# Incident: [short name] — [SEV1 / SEV2 / SEV3]
**Comms owner:** [you] · **Incident commander:** [name] · **Status:** [Investigating / Identified / Mitigating / Monitoring / Resolved]
**Started:** [time TZ] · **Detected:** [time TZ] · **Next update:** [time TZ]

---

## 1. Internal escalation note (Slack / email to leadership + support)
**🔴 [SEV_] [Product/feature] [is down / is degraded] — [one-line symptom]**
- **Impact:** [who/what is affected — segment, %, surface]. [Is data/revenue/SLA at risk? y/n + detail.]
- **Started / detected:** [time] / [time TZ]. **Duration so far:** [Xm].
- **Status:** [Investigating / Identified / Mitigating / Monitoring]. **Current theory:** [what we suspect — label as unconfirmed].
- **What we're doing:** [the active step].
- **On it:** IC [name] · Eng [name(s)] · Comms [you].
- **For support — holding line, read verbatim:** "[one self-contained sentence support can paste to any inbound — what's affected, that data is safe if true, and when we'll next update]." Do not speculate on cause or ETA.
- **Ask of leadership:** [decision needed, or "none — informational, next update [time]"].

## 2. Customer-facing status update (status page / in-app banner / email)
> Post the first one fast; repeat on cadence until resolved. No internal names, no cause speculation, no ETA you can't keep.

**[Investigating] [Product] — [plain-language symptom]**
[Timestamp TZ] — We're aware that [what users are experiencing, in their words]. [Who/what is affected, if it's a subset.] Our team is actively investigating. We'll post another update by **[next update time TZ]**.

**[Identified]** [time TZ] — We've identified the cause and are working on a fix. [Affected functionality] may still be [symptom]. Next update by **[time TZ]**.
**[Monitoring]** [time TZ] — A fix is deployed and [functionality] is recovering. We're monitoring to confirm full recovery. Next update by **[time TZ]**.

## 3. Resolution / all-clear
**[Resolved] [Product] — service restored**
[time TZ] — This is resolved. [What was affected] is fully operational as of **[time TZ]**, after [duration] of impact. [Residual note, e.g., "Delayed [data] is backfilling and will be current within [X]."] [Any user action needed, or "No action needed on your part."]
We're sorry for the disruption[ and for any [specific cost to them, e.g., missed digests]]. [For SEV1/2: A full postmortem will follow by [date].]

## 4. Blameless postmortem (for PM / leadership)
**Incident:** [name] · **Severity:** [SEV_] · **Duration:** [start–end TZ, total Xh Ym] · **Author:** [you] · **Date:** [date]

**TL;DR:** [1-2 sentences: what broke, who it hit, how long, and that it's resolved. A reader who stops here knows the impact.]

**Customer impact**
- **Who:** [segment / count / %]. **What they experienced:** [symptom]. **Business impact:** [SLA credits, churn risk, revenue, support volume — quantify if you can].

**Timeline** (all [TZ])
| Time | Event |
|------|-------|
| [t0] | [trigger — e.g., "Config change deployed"] |
| [t1] | [first user impact] |
| [t2] | [detected — by alert or customer report? note which] |
| [t3] | [mitigated] |
| [t4] | [resolved] |

**Root cause (plain language):** [the *system/process* reason, one or two sentences — no individual named]. [Point to the eng RCA for technical depth.]

**What went well / what hurt**
- ✅ [e.g., "Alert fired in 4 min."]  ⚠️ [e.g., "No customer-facing status update for the first 40 min."]

**Prevention — action items**
| Action | Type (Detect / Prevent / Mitigate) | Owner | Due |
|--------|-----------------------------------|-------|-----|
| [specific change] | Prevent | [name] | [date] |
| [specific change] | Detect | [name] | [date] |

Avoid (anti-patterns)

• Inventing an ETA to calm people down. "Fixed within the hour" that slips destroys more trust than the outage. Promise the next update time, never the fix time, unless eng has truly committed.
• Going silent between updates. A missed cadence reads as "they've lost control." Ship the on-time update even when it's "still investigating, next update [time]."
• Naming a person or guessing the cause in customer comms. "A bad deploy by [name]" or an unconfirmed theory in public is a credibility (and HR) grenade. Cause talk is blameless, internal, and confirmed-only.
• Copy-pasting the internal note to customers. Internal carries scope %, theories, and hostnames; customers get impact + action + next update. One leak of an internal note becomes a tweet.
• A postmortem with no owners or dates. "We should add monitoring" with nobody on the hook is a wish. Every action item gets a name and a date or it didn't happen.
• Apologizing before you've resolved it. A grand apology mid-incident reads as panic; the apology belongs in the all-clear, sized to the actual impact.
• Declaring "resolved" on hope, not monitoring. An all-clear that bounces back into a second incident burns trust twice — confirm recovery on the dashboard before you post it.

Quality bar

Before delivering, verify every box:

• [ ] Severity is set, and the matching cadence (SEV1→30m / SEV2→60m / SEV3→status-change) is committed in the first message.
• [ ] Every customer-facing message ends with a concrete next-update time — never a fix time you can't stand behind.
• [ ] The internal note and the customer note are genuinely different: no scope %, theory, hostname, or person's name leaked into anything customer-facing.
• [ ] No invented ETA, root cause, or customer count anywhere — gaps are stated as "we don't yet know X."
• [ ] The support holding line is self-contained and verbatim-safe (a support rep could paste it without asking you a question).
• [ ] The all-clear is sent only after monitoring confirms recovery, and includes residual effects + any user action.
• [ ] The postmortem leads with a TL;DR a leader can stop reading after, and states impact in customer/business terms.
• [ ] The postmortem root cause names a system/process, not a person — and points to the eng RCA rather than reproducing it.
• [ ] Every prevention action item has an owner and a due date.
• [ ] One consistent, labeled timezone across all artifacts.

Tips

• First message in 5 minutes beats a perfect one in 30. "We're aware and investigating, update by [time]" is enough — acknowledgement is the product early on.
• Severity sets cadence, cadence sets trust. Decide SEV first; everything (update frequency, who you wake up, customer tone) falls out of it.
• Write the next-update timestamp into every message. It converts "are they even looking?" anxiety into a wait with a deadline — and forces you to actually post.
• Separate the comms owner from the incident commander. The person fixing it should not be the person tweeting; split the roles even on a small team so neither job starves.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Sidsaladi9
• Source: Sidsaladi9/persona-os
• License: MIT
• Homepage: https://sidsaladi.substack.com

Install and usage instructions live in the source repository linked above.