groom-milestone

Groom Milestone

Prepare a GitHub milestone for parallel execution by /work-milestone.

Outcome

All items in the milestone are groomed, dependency-analyzed, and conflict-grouped — producing a dispatch plan that /work-milestone can execute without further human input except blocker resolution.

Entry Conditions

• Milestone number provided as first argument
• Milestone exists on GitHub with state=open
• At least one item assigned to the milestone (via /group-items-to-milestone)
• Backlog MCP server responding

Exit Conditions

• Every item in the milestone has groomed: true
• A priority-ordered list with wave assignments is produced
• A dependency graph (item-to-item) is produced with conflict groups identified
• Any items recommended for splitting have been split (user-approved)
• Any items recommended for addition have been added (user-approved)
• Dispatch plan stored via dispatch_create_plan(milestone_number={N}, ...) and retrievable with dispatch_read({N})

Workflow

flowchart TD
    Start(["Input: milestone number"]) --> Load["Step 1: Load Milestone<br>Action: backlog_list_issues(milestone=N)<br>Output: list of items with groomed status"]

    Load --> Assess["Step 2: Assess Milestone Scope<br>Action: For each item, read title + description + labels<br>Output: scope summary — what the milestone covers"]

    Assess --> GapCheck{"Step 3: Gap Analysis<br>Areas the milestone should cover<br>but no item addresses?<br>Observable: compare milestone title/description<br>against item titles and descriptions"}

    GapCheck -->|"Gaps found"| ProposeAdd["Step 3a: Propose Additions<br>Action: Present gaps to user with<br>suggested new items or existing<br>backlog items to associate<br>Output: user decision per gap"]
    GapCheck -->|"No gaps"| GroomGate

    ProposeAdd --> AddItems["Step 3b: Execute Additions<br>Action: create-backlog-item, then<br>github_project_setup.py issue set-milestone<br>to assign it (see MCP Tools Used)<br>Output: updated milestone item list"]
    AddItems --> GroomGate

    GroomGate{"Step 4: Groom Check<br>Any items with groomed=false?<br>Observable: groomed field in<br>backlog_list_issues response"}

    GroomGate -->|"Ungroomed items exist"| BatchGroom["Step 4a: Batch Groom via Kage-Bunshin<br>Actor: Parallel kage-bunshin sessions<br>Action: For each ungroomed item, spawn:<br>claude -p --model sonnet --permission-mode auto<br>--output-format json --no-session-persistence<br>'Load /dh:groom-backlog-item {title}'<br>All sessions run in same directory (no worktree —<br>grooming writes go through backlog MCP, not filesystem).<br>Wait for all PIDs to exit."]
    GroomGate -->|"All groomed"| DepAnalysis

    BatchGroom --> GroomResults{"Step 4b: Check Groom Results<br>Read each /tmp/kb-groom-{issue}.json<br>Any sessions failed?"}
    GroomResults -->|"All succeeded"| DepAnalysis
    GroomResults -->|"Some failed"| GroomFix{"Fixable without<br>user direction?"}
    GroomFix -->|"Yes"| GroomRetry["Re-spawn failed items only"]
    GroomRetry --> GroomResults
    GroomFix -->|"No"| GroomEscalate["Report to user:<br>item, error, what was attempted"]
    GroomEscalate --> GroomUserQ{"User says?"}
    GroomUserQ -->|"Skip items"| DepAnalysis
    GroomUserQ -->|"Abort"| Abort(["ABORT — user decision"])

    DepAnalysis["Step 5: Dependency Analysis<br>Action: Read Impact Radius from each groomed item.<br>Compare file lists across all items to find overlaps.<br>Output: dependency graph — which items<br>touch overlapping files/modules"]

    DepAnalysis --> ConflictGroup["Step 6: Conflict Grouping<br>Action: Items with file overlap form a conflict group.<br>Items in the same conflict group MUST execute sequentially.<br>Items in different groups or with no overlap execute in parallel.<br>Output: conflict groups list"]

    ConflictGroup --> SplitCheck{"Step 7: Split Assessment<br>Any single item spanning multiple<br>independent plugins or repo areas?<br>Observable: item's Impact Radius<br>lists files in 2+ unrelated directories"}

    SplitCheck -->|"Splittable items found"| ProposeSplit["Step 7a: Propose Splits<br>Action: Present split recommendations<br>to user with rationale per item<br>Output: user decision per split"]
    SplitCheck -->|"No splits needed"| Prioritize

    ProposeSplit --> ExecSplit["Step 7b: Execute Splits<br>Action: Create new backlog items for each split piece.<br>Assign to this milestone. Close original if fully decomposed.<br>Output: updated item list"]
    ExecSplit --> DepAnalysis

    Prioritize["Step 8: Priority Ordering<br>Action: Order items by:<br>1. Dependency constraints (blocked-by first)<br>2. Priority label (P0 > P1 > P2)<br>3. Conflict group (parallel-safe first)<br>Output: ordered list with wave assignments"]

    Prioritize --> WavePlan["Step 9: Build Dispatch Plan<br>Action: Assign items to waves.<br>Wave 1: all items with no dependencies and no conflict group overlap.<br>Wave 2: items unblocked after Wave 1. Continue until all items assigned.<br>Verify wave ordering and dependency references.<br>Construct DispatchPlan object from wave assignments.<br>Call dispatch_create_plan(milestone_number=N, plan=dispatch_plan) to persist plan atomically.<br>For re-grooming a stale plan, pass overwrite=True.<br>Call dispatch_wave_start per wave to register state.<br>Output: plan/milestone-{N}-dispatch.yaml"]

    WavePlan --> Report["Step 10: Report<br>Output: milestone summary with wave assignments,<br>conflict groups, estimated parallelism per wave,<br>and next command: /work-milestone {N}"]

    Report --> Done(["Exit: Dispatch plan ready"])