| name | retrospective |
| description | Extract actionable learning from completed work — what went well, what didn't, and what will actually change as a result. Use after a project, sprint, release, or significant event. Produces concrete action items, not just discussion. Trigger phrases: "retrospective", "retro", "what did we learn from this project?", "what went wrong?", "how do we improve?", "post-project review", "what should we do differently next time?", "team reflection", "lessons learned". A retrospective without action items is a venting session. This skill produces change. DISAMBIGUATION — "what went wrong" is ambiguous: learning from a COMPLETED project or sprint to improve future work → retrospective (this skill); finding the technical cause of a SPECIFIC failure or bug → debugging-rootcause; deciding the future DIRECTION of an ongoing project → project-compass. Choose retrospective when the work is done and the goal is team learning.
|
Retrospective Skill
Converts experience into improvement. A retrospective done well changes
how the team works next time. A retrospective done poorly wastes an hour.
Core Principle
The goal is not to discuss the past — it is to change the future.
Every retrospective must end with specific, owned, time-bounded actions.
If nothing changes after the retrospective, it was not worth having.
Before the Retrospective
□ Gather data (don't rely on memory alone):
- Timeline of key events
- Metrics (velocity, quality, incidents, schedule variance)
- Feedback from stakeholders or users if available
□ Create psychological safety:
- Blameless framing: "how did our systems and processes fail?" not "who failed?"
- Share data before opinions — ground discussion in facts
□ Set time box: 60-90 minutes for a sprint; up to 2 hours for a major project
□ Identify a facilitator (not the team lead — they should participate freely)
Core Format: Start / Stop / Continue
Simple, works for most teams.
START: What should we begin doing that we aren't doing?
STOP: What should we stop doing that isn't serving us?
CONTINUE: What is working well that we should keep doing?
Process:
1. Silent individual writing (5-7 minutes): everyone writes their own items
2. Share and cluster: group similar items together
3. Dot voting: each person gets 5 votes to prioritize (can stack on one item)
4. Discuss top-voted items (time-boxed: 5-7 minutes per item)
5. Convert to action items
Format for Project-Level Retros: 4Ls
For larger projects or major releases where more depth is needed.
LIKED: What went well? What would we do again?
LEARNED: What did we discover? What surprised us?
LACKED: What was missing? What would have helped?
LONGED FOR: What do we wish we had? What would we want next time?
Process: same as Start/Stop/Continue — write, share, cluster, vote, discuss, action
Format for Incident or Failure Retros
For retrospectives focused on something that went wrong.
TIMELINE: Walk through what happened (from the scribe's log if available)
→ Makes the discussion concrete, not memory-based
WHAT WENT WELL:
Things that worked during the incident/failure response.
Don't skip this — it prevents over-correcting and recognizes good work.
WHAT COULD BE IMPROVED:
Systems, processes, tools, communication — not people.
"Our monitoring didn't alert us" not "Alex didn't notice"
5 WHYS:
Drive to root cause on the most important failure.
Stops at the level where a systemic change is possible.
ACTION ITEMS:
Specific, owned, time-bounded. See action item format below.
Converting Discussion into Action Items
This is where most retrospectives fail — vague commitments that never happen.
BAD ACTION ITEM:
"Improve our testing"
"Better communication"
"Document things more"
GOOD ACTION ITEM:
"Add integration tests for the payment flow — Owner: Sam — By: Friday"
"Send a weekly project status email every Monday — Owner: Jordan — Starting: next week"
"Write the API documentation for the 3 new endpoints — Owner: Alex — By: next sprint"
GOOD ACTION ITEM CRITERIA:
□ Specific: anyone reading it knows exactly what to do
□ Owned: one named person is responsible (not "the team")
□ Time-bounded: a specific date or milestone, not "eventually"
□ Achievable: can realistically be done in the timeframe
ACTION ITEM LIMIT:
Maximum 3-5 action items per retrospective.
Teams that generate 15 action items complete 0.
Choose the 3-5 that will have the most impact.
Retrospective Output Format
RETROSPECTIVE RECORD
══════════════════════════════════════
Project/Sprint: <name>
Date: <date>
Participants: <names>
Facilitator: <name>
WHAT WENT WELL:
+ <item>
+ <item>
WHAT COULD BE IMPROVED:
- <item>
- <item>
KEY INSIGHTS:
<1-3 most important observations from the discussion>
ACTION ITEMS:
□ <specific action> — Owner: <name> — By: <date>
□ <specific action> — Owner: <name> — By: <date>
□ <specific action> — Owner: <name> — By: <date>
REVIEW DATE: <when action items will be checked>
══════════════════════════════════════
Making Retrospectives Stick
The most common failure: action items are written and never reviewed.
□ Assign a single owner to each action item (not "the team")
□ Schedule the review date before leaving the retrospective
□ Open the NEXT retrospective by reviewing last time's action items
□ If an action item was not completed: why not? What's the blocker?
RETROSPECTIVE HEALTH CHECK (ask periodically):
"Are things actually changing as a result of our retrospectives?"
YES → keep the format; consider whether to increase frequency
NO → something is wrong: safety? action quality? follow-through?
address the meta-problem before running more retrospectives