Turn incident lessons into updated runbooks
After the post-mortem doc ships ([Run a post-mortem that leads to action](/tutorials/run-a-post-mortem-that-leads-to-action)) — **promote follow-ups into Auxot** — context-file paragraphs, agent instruction patches, workflow checklist rows, and new regression prompts ([Catch regressions after you change an agent](/tutorials/catch-regressions-after-you-change-an-agent)) — so *what we learned* lives where operators actually work, not only in a PDF.
Plus: three Admin-Agent passes — map pasted action items → Auxot surface each belongs on, draft **Settings → Context Files** patch notes + [Connect Discord to your agents](/tutorials/connect-discord-to-your-agents) changelog bullets, and propose one **scheduled canary** ([Run scheduled canary checks on production agents](/tutorials/run-scheduled-canary-checks-on-production-agents)) tuned to the failure class.
| Audience | Developers · Admins · Everyone |
|---|---|
| Time | ~12 min |
| Prerequisites | Factual review habit ([Run a post-mortem that leads to action](/tutorials/run-a-post-mortem-that-leads-to-action)). Somewhere durable for prose ([Add your first context file](/tutorials/add-your-first-context-file)). Helpful: announcing agent updates ([Update your agents without breaking the team](/tutorials/update-your-agents-without-breaking-the-team)), regression packs ([Catch regressions after you change an agent](/tutorials/catch-regressions-after-you-change-an-agent)), quarterly roster rhythm ([Run a quarterly review of your agents](/tutorials/run-a-quarterly-review-of-your-agents)). |
| You'll end up with | A **runbook promotion checklist** tied to one real incident — each follow-up row shows **Auxot destination** (file, agent, or workflow), **patch stub**, **owner**, and **done date** — Internal Q&A updated when the lesson was *how do we…* ([Build an internal Q&A agent for your team](/tutorials/build-an-internal-q-and-a-agent)). |
When a tutorial shows italic text in quotation marks, it usually mirrors a label or helper string inside Auxot. Product copy changes between releases — if something reads differently in your workspace, trust what you see on screen.
Callouts with a Worth knowing gold accent are meant as must-read context before you move on. Blockquotes that open with Tip are lighter, optional depth.
Why this matters
Post-mortems die in shared drives. The team nodded in the meeting; then nobody edits the on-call doc, nobody tightens the agent that walks newcomers through the fragile path, and the same surprise repeats eight months later.
Auxot is where your operators already ask questions. Runbooks here mean:
- Context files: facts, limits, never do X after deploy (Add your first context file).
- Agent instructions: refusal language, escalation cues (Give your agent its job description).
- Workflows: human columns that force a check (Run a workflow).
- Regression + canary prompts: prove the hole stays plugged (Catch regressions after you change an agent, Run scheduled canary checks on production agents).
This lesson is the bridge from Run a post-mortem that leads to action’s narrative to editable artifacts, still human-driven: you paste follow-ups, you approve diffs, you tell the org via change notes (Update your agents without breaking the team).
Nothing updates on its own — you patch files, you assign owners, you verify Jobs after reruns (View your audit logs).
Quick start
- Export action items — from the post-mortem agent’s output or tracker: one bullet per row, owner attached (Run a post-mortem that leads to action).
- Classify destination — label each: Context, Agent prose, Workflow, Tool/policy, Regression/canary, or Comms only; ambiguous rows stay human ticket, not forced into Auxot.
- Draft patches — paste batch into Admin Agent (power move 1): get suggested markdown inserts; you still edit Settings.
- Ship change notes — anything teammates notice in Chat gets a dated line (Update your agents without breaking the team).
- Verify — rerun affected regression rows (Catch regressions after you change an agent): optional schedule new canary (Run scheduled canary checks on production agents); confirm green Jobs.
Done? Checklist mostly complete or explicitly deferred; retro link referenced in quarterly roster review appendix (Run a quarterly review of your agents).
The agent can do that?
1. Destination mapping
Chat → Admin Agent:
Post-mortem action items pasted: […]. Table: Item | Auxot destination (context/agent/workflow/policy/regression/none) | One-line patch idea | Risk if skipped — markdown — no implementation yet.
Why it’s non-obvious: Meetings produce vague monitor better rows; mapping forces concrete surfaces after you paste text.
2. Context-file patch draft
Target context file "[title]" — incident lesson: [two sentences]. Draft markdown subsection **Incident YYYY-MM — [codename]** — bullets only — <=120 words — cite retro link placeholder.
Why it’s non-obvious: Raw lessons ramble; constrained subsection lands in Settings → Context Files without overwriting unrelated guidance until you paste.
3. Canary / regression suggestion
Failure class: [provider timeout | bad prompt | missing human gate]. Recommend add regression row vs weekly canary ([Catch regressions after you change an agent](/tutorials/catch-regressions-after-you-change-an-agent) vs [Run scheduled canary checks on production agents](/tutorials/run-scheduled-canary-checks-on-production-agents)) — pick one — justify in two sentences.
Why it’s non-obvious: Teams file tickets instead of monitors; explicit pick prevents both everywhere after you describe failure.
Go deeper
Internal Q&A
When the gap was how do we…?, append handbook-style answer Internal Q&A cites (Build an internal Q&A agent for your team); still owned by docs maintainers.
Pre-ship reviews
Fold repeating incident themes into pre-ship reviews (Run a release readiness review): narrative linkage beats silent hope.
Slack timelines
Incident channels often hold facts worth excerpting (Connect Slack to your agents): redact customer identifiers before pasting into org-wide context.
Adversarial follow-up
If incident involved adversarial misuse, extend adversarial tests (Red-team your agents against prompt injection) alongside benign regression rows.
Walkthrough
Step 1: Link retro doc
Ticket URL or wiki slug; every checklist row references it.
Step 2: Edit context files first
Facts agents consume beat announcement emails.
Step 3: Patch agents + workflows
Settings → Agents instructions; workflow columns per Run a workflow.
Step 4: Extend regression pack
New rows: Last green reset (Catch regressions after you change an agent).
Step 5: Tell humans
Directory or Slack post (Build your agent directory, Connect Discord to your agents): what changed + why + where to read.
What’s next
- → Run a post-mortem that leads to action. Narrative upstream; this lesson is promotion downstream.
- → Update your agents without breaking the team. Human-visible deltas ship with patches.
- → Catch regressions after you change an agent. Encode lessons as prompts you rerun on purpose.
- → Run scheduled canary checks on production agents. Time-based variance watch when incidents were silent until late.
- → Turn account research into CRM field updates. Promote crisp facts into systems of record; same “patch what operators touch” habit outside the incident doc.
- → Build an internal Q&A agent for your team. Operational FAQs belong beside policies.
Reference
- Pages in Auxot: Settings → Context Files, Settings → Agents, Workflows, and Audit Logs → Jobs
- See also: Triage pull requests before humans review them, Turn account research into CRM field updates, View your audit logs, Run a quarterly review of your agents, Define a tool policy