---
name: bmad-retrospective
description: Derive specific improvement actions from a completed delivery interval and its evidence.
---

# Retrospective

Read the [execution guide](../../shared/execution.md) and
[Oholiab's role](dev-agent.md).

## Inputs

At least one explicit delivery artifact, with the epic, sprint or interval to
review identified in the artifact or request. Inspect available plans, delivery
reports, issues and check results. The absence of participant feedback limits
what can be said about people's experiences.

## Procedure

1. Define the interval and intended outcomes. Establish what was actually
   delivered, deferred or left unverified from the available evidence.
2. Build a short timeline of relevant events, decisions and observed effects.
   Cite the source of each material event and separate observation from an
   inferred explanation.
3. Examine what helped delivery and where rework, delay or defects occurred.
   Compare outcomes with the stated goal and constraints. Avoid attributing
   motives or blame to people from incomplete project records.
4. Investigate plausible causes using supporting and contrary evidence. A
   correlation or one anecdote is a hypothesis until more evidence supports it.
   Identify decisions that were reasonable given what was known at the time.
5. Select a small set of concrete improvements proportional to the observed
   problem. Give each an intended result, proposed owner or unassigned state,
   completion evidence and a point for checking whether it helped.
6. Compare with earlier retrospective actions if available. Record actions that
   worked, remain open or should change. Do not create repeated action items that
   ignore the existing follow-up record.
7. Write the retrospective and distinguish accepted project decisions from
   recommendations. Proposed ownership does not assign work or notify anyone.

## Output and acceptance

Write the report for retrospective with scope, outcome comparison, evidence
timeline, supported lessons, unresolved hypotheses and an action table.

| Improvement | Supporting evidence | Intended result | Owner or unassigned | Follow-up check |
| --- | --- | --- | --- | --- |
| Concrete change | Source observation | Observable benefit | Actual agreement or proposal | Measurement and review point |

Each proposed action should address an evidenced problem or reinforce an evidenced
success. Zero new actions is valid when the reviewed evidence does not justify
changes. Do not claim team consensus from a single assistant's analysis.

## Continue

Preserve participant additions and prior decisions. Revisit only conclusions
affected by new evidence, then update action outcomes with actual observations.
Keep a hypothesis provisional when the expected follow-up evidence is absent.