sprint-execution

flowchart TD
    A([BEGIN]) --> B[Read sprint goals from .ai/instructions/]
    B --> B2[Check past lessons in .ai/lessons/]
    B2 --> C[Dispatch Claude Code to decompose into tasks]
    C --> D[Review task briefs for completeness]
    D --> E{Are all tasks well-defined?}
    E -->|No| F[Request Claude Code to refine unclear tasks]
    F --> D
    E -->|Yes| G[Assign first unblocked task to appropriate agent via .ai/instructions/]
    G --> H[Wait for agent submit commit]
    H --> I[Review submitted work against acceptance criteria]
    I --> J{Work approved?}
    J -->|No| K[Write feedback to .ai/reviews/ and commit reject]
    K --> L[Agent addresses feedback on their branch]
    L --> H
    J -->|Yes| M[Merge agent branch to pre-mortal]
    M --> N[Update .ai/status.md with task completion]
    N --> O{More tasks remaining?}
    O -->|Yes| G
    O -->|No| P[Generate sprint summary report in .ai/reports/]
    P --> Q[Notify Human PM that pre-mortal is ready for review]
    Q --> R([END])

Flow Steps Detail

Step: Read sprint goals

• •Read .ai/instructions/ for the current sprint brief
• •Identify sprint scope, priorities, and constraints

Step: Check past lessons (before decomposition)

• •Read .ai/lessons/applied-lessons.md for patterns already applied to this project
• •If similar work was done in a past configuration, reference those patterns:
  • •Check task sizing (past configs show what worked for single-session scope)
  • •Check dependency patterns (past configs show what caused blocking)
  • •Check acceptance criteria format (past configs show what was testable)
• •Avoid patterns marked as "failed" or "anti-pattern" in .ai/lessons/
• •Apply patterns marked as "successful" to the current decomposition
• •If no lessons file exists, run ./scripts/extract-past-lessons.sh first

Step: Dispatch Claude Code to decompose

• •Create instruction for Claude Code in .ai/instructions/claude-decompose-sprint-X.md
• •Claude Code writes task briefs to .ai/tasks/

Step: Review task briefs

• •Check each task has: Status, Priority, Type, Dependencies, Acceptance Criteria
• •Verify file ownership matches agent assignment
• •Ensure no circular dependencies

Step: Assign task to agent

• •Pick the highest-priority unblocked task
• •Write instruction to .ai/instructions/<agent>-<task>.md
• •Commit with [AGENT:kimi] [ACTION:delegate] [TASK:X]

Step: Review submitted work

• •Read the task brief acceptance criteria
• •Run git diff on the agent's branch
• •Check boundary compliance
• •Check for regressions
• •Use the code-review flow for detailed review

Step: Merge to pre-mortal

• •git merge <agent>/<task-id> --no-ff
• •Commit with [AGENT:kimi] [ACTION:merge] [TASK:X]

Step: Generate sprint summary

• •Count tasks completed, blocked, review cycles
• •Write report to .ai/reports/sprint-X-summary.md
• •Commit with [AGENT:kimi] [ACTION:report] [SPRINT:X]