agentsclimarketplace

Postmortem review board

Skill Amey-Thakur/AI-SKILLS/skills/big-tech-processes/postmortem-review-board

Plug-and-play skills and prompts for every AI coding agent

Install
npx -y skills add Amey-Thakur/AI-SKILLS --skill postmortem-review-board

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 19 days oldThe repository was created 19 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Run an org-level board that grades incident postmortems against a quality bar, promotes cross-cutting fixes to owned work, and reads across incidents for trends. Use when your organization writes enough postmortems that reviewing them one at a time hides the systemic pattern.

SKILL.md

3.7 KB, as published. Nobody here has run it

Postmortem review board

One postmortem closes one incident. A board exists to see what a single document cannot: the third outage this quarter with the same trigger, the action items filed and never shipped, the failure class spreading quietly between teams. Without it, each team learns its lesson alone and the organization pays for the same lesson again at a higher price.

Method

  1. Gate entry on severity and impact, never on volunteering. Every SEV1 and SEV2, plus any incident above a stated customer or revenue threshold, comes to the board. Route lower severities to team-local review so the board reads the incidents that carry org-wide signal instead of drowning in all of them.
  2. Grade against a written rubric before the discussion starts. A pass needs a real timeline, a root cause that survives "five whys" past the first human in the chain, contributing factors, and action items with named owners and dates. "Operator error" is where the analysis quit, not a root cause. Bounce failing writeups instead of workshopping them live.
  3. Keep it blameless, out loud and every time. The board judges the system and the writeup, never the responder. The first review that turns into a tribunal is the last honest postmortem you get: the next one arrives sanitized, and sanitized postmortems are the board's blind spot.
  4. Split team-local fixes from cross-cutting actions. For each incident, ask whether the remedy protects only the team that got paged or a weakness others share. Promote the shared ones to tracked, owned work with an executive sponsor: a retry storm that sank one service will sink the next until the common client library changes.
  5. Age the action items and refuse to let them rot. Report open action-item count and age at every session. Postmortem follow-ups are the first work dropped when a team gets busy, so a board that files them without tracking closure is theater. A recurring incident whose prior fix is still open is the board's sharpest indictment of itself.
  6. Read across the corpus, not just down each report. Tag every incident by trigger class: deploy, config change, capacity, dependency failure, expiry. Each quarter, name the top classes and the teams they cluster in. That cross-sectional pattern is invisible in any single document and is the board's real output.
  7. Route the trends into planning. Convert the recurring classes into roadmap input: a hardening project, a guardrail, a paved path off the sharp edge. A trend report that reaches no planning cycle changes nothing next quarter, and the board decays into a reading group.

Signals

  • Would a postmortem that blamed a person and stopped there get sent back, or wave through on a busy day?
  • Can you state this quarter's top three trigger classes and who owns reducing each?
  • Is open action-item age trending down, or does the backlog only accumulate?
  • When an incident recurs, does the board notice that its prior fix never shipped?

Boundaries

This is the org-level review of many postmortems, not the authoring of one: writing an individual incident report belongs to your incident-response process. The board sets a quality bar and finds patterns; it does not run the live incident or approve releases. Match cadence and severity thresholds to your existing incident taxonomy rather than inventing a parallel scale.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.