Skip to main content

metrics.sprint-retro

Type: commands | Track: Enterprise

Sprint retrospective ceremony. Collects DORA actuals, agent consensus scores, 4L review (Liked, Learned, Lacked, Longed For), and writes retro evidence. Autonomous local-first. Delegates to observability-engineer. Exempt from PO+CA coordination.


Quick Start

# Run sprint retro
/metrics:sprint-retro

# Override sprint
/metrics:sprint-retro --sprint CloudOps-S1

Output Sections

SectionContent
DORA actualsDF, LT, CFR, MTTR — actual values with target comparison
4L ReviewLiked / Learned / Lacked / Longed For
Agent consensusPO + CA + QA scoring round summary
ACT itemsImprovement actions with owner and sprint assignment
EvidenceJSON retro file written to framework/retrospectives/

Benefits

  1. DORA actuals sourced from dora.csv — no manual metric collection
  2. Agent consensus section captures scoring round outcomes for governance audit trail
  3. ACT items written with owner and sprint — prevent DEFERRED_GAP_LOOP where retro findings are never executed
  4. Evidence written to framework/retrospectives/ — git-tracked for continuous improvement audit

When to Use

AttributeDetail
PersonaHITL / Engineering Manager
TriggerSprint end (bi-weekly) — after the sprint review, when the team needs to reflect on HOW they performed and capture improvement actions with owners
Business Value4L review, DORA actuals, and improvement actions with owners — gives the team an objective retrospective with actionable outcomes, not just a feelings-based discussion
FrequencyBi-weekly

Example: As an Engineering Manager (HITL), I need the sprint retrospective output because I need to capture what worked, what didn't, and what we'll change, with specific ACT items assigned to owners before the next sprint starts. I run /metrics:sprint-retro which produces the 4L review from coordination logs, DORA actuals vs targets, agent consensus summary, and ACT items written to framework/retrospectives/sprint-retro-{sprint}-{date}.md.

Coordination-exempt command. Reads existing data only — no architecture decisions.