Execution Retrospective

Execution Retrospective analyzes one completed, interrupted, or failed Moira execution from authorized execution context, agent-session evidence, artifacts, reviews, retries, user corrections, and terminal behavior. It produces prioritized proposals for improving the result, workflow definition, or system prompt without changing the subject execution.

mcp__moira__start({
  action: "prepare",
  workflowId: "moira/execution-retrospective",
  parentExecutionId: "none"
})
mcp__moira__start({ action: "execute", startAttemptId: "<Start attempt ID from prepare>" })

Give the execution to analyze at intake. Its identity is separate from the parent: a completed execution cannot be a parent. Use "none" for a standalone retrospective, or the actual running caller’s execution ID when it starts the retrospective as a child.

Evidence and process

The workflow materializes an execution-correlated workspace and preserves whether evidence is sufficient, partial, or unavailable. Independent analysis separates observed facts from inference, explains what worked, identifies causes of rework or delay, and writes a source-linked retrospective. Independent final review must clear current blockers; confirmed report findings use contained or spreading repair routes.

Partial or unavailable execution evidence cannot become a confident diagnosis. The terminal result remains truthful about the evidence boundary and points to the reviewed proposal.

Authority

The workflow is proposal-only. It does not edit the analyzed execution, mutate a workflow or prompt, publish, notify, commit, release, deploy, or change production. Use Workflow Management Flow separately to implement accepted workflow changes.