Agents / Skills

Retrospective Format

Skill

Designs the retrospective your team actually needs this time: the format fitted to what the period was like, the discussion structured past the usual suspects, and the output filtered to the few actions that will genuinely happen — breaking the groundhog-day retro cycle.

Best for

  • Retros that produce the same three complaints and zero change
  • The after-something-big review (launch, incident, quarter) that deserves better than 'what went well'
  • Teams who stopped believing in retros because nothing ever came of them

What you give it

  • What the period was like (the rough shape: a grind, a win, a mess), the team's mood, and what past retros produced

What you get back

  • The format fitted to the moment: the structure, the prompts and the timeboxes — different for the post-incident, the post-win and the stuck-team retro, because one format for all moments is why retros go stale
  • The discussion mechanics that surface what the round-robin misses: silent writing before speaking, the quiet voices drawn in, the recurring complaint traced to its system instead of re-aired
  • The action filter: few, owned, dated, checkable — and the previous retro's actions reviewed FIRST, because that accountability loop is what makes the whole ritual real

How it works

  1. Fits the format to the moment: the post-incident wants a timeline walk, the post-win wants to extract the repeatable, the grind wants energy diagnosis, the stuck team wants the stale complaint finally traced — the all-purpose format is why retros produce all-purpose nothing.
  2. Structures past the loudest voices: silent writing first, grouped themes, dot-votes where breadth matters — the mechanics that get the real observations into the room before the usual narrative takes it.
  3. Traces instead of re-airs: the recurring complaint gets the five-whys treatment to a system cause with a fixable handle, because its sixth appearance as a vibe is the format's failure, not the team's.
  4. Filters the actions brutally: one to three, each owned, dated and checkable — and opens every retro with the previous actions' fates, the accountability loop that separates ritual from theatre.

Example

You: Our retros have produced 'communication could be better' for six sprints. Design the next one.

Result: The design: previous actions reviewed first (the ritual's missing spine — four of six past actions had silently died; naming that was the retro's real opening), silent writing before any speaking (the two quiet engineers produced the sharpest observations — the round-robin had been letting the loud process their thoughts aloud), the stale complaint finally traced instead of re-aired ('communication could be better' decomposed with five-whys into: the handoff between design and build has no definition of ready — a fixable system, not a vibe), and the action filter applied ruthlessly: ONE action (the definition-of-ready, drafted by Thursday, owner named), because six sprints of three-actions-each had produced eighteen orphans. Next retro opened with: the action happened, the handoff complaints stopped. The retro after that had attendance without reminders.

Limits — please read

  • Format cannot fix safety: if people cannot speak freely, that is the prior problem, and the design flags the signs rather than pretending prompts solve them.
  • The facilitation matters; the design arms whoever runs it with the timeboxes and the interventions, but someone must hold them.
  • Systemic causes beyond the team's control get routed (to the stakeholder update, the escalation) rather than becoming the annual complaint.