Paths: File paths (shared/, references/, ../ln-*) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root.

Story Execution Orchestrator

Executes a Story end-to-end by looping through its tasks in priority order. Sets Story to To Review when all tasks Done (quality gate decides Done).

Purpose & Scope

• Load Story + task metadata (no descriptions) and drive execution
• Process tasks in order: To Review → To Rework → Todo (foundation-first within each status)
• Delegate per task type to appropriate workers (see Worker Invocation table)
• Mandatory immediate review: Every execution/rework → ln-402 immediately. No batching

Task Storage Mode

MANDATORY READ: Load shared/references/storage_mode_detection.md for Linear vs File mode detection and operations.

When to Use

• Story is Todo or In Progress and has implementation/refactor/test tasks to finish
• Need automated orchestration through To Review

Workflow

Phase 1: Discovery & Branch Setup

1. Auto-discover Team ID/config from kanban_board.md + CLAUDE.md
2. Get Story identifier (e.g., PROJ-42) and title
3. Generate branch name: feature/{identifier}-{story-title-slug} (lowercase, spaces→dashes, no special chars)
4. Check current branch: git branch --show-current
5. If not matching:
   • If branch exists: git checkout feature/{identifier}-{slug}
   • If not: git checkout -b feature/{identifier}-{slug}

Phase 2: Load Metadata

Fetch Story metadata and all child task metadata (ID/title/status/labels only):

• Linear Mode: list_issues(parentId=Story.id)
• File Mode: Glob("docs/tasks/epics/*/stories/{story-slug}/tasks/*.md") + parse **Status:**

Summarize counts (e.g., “2 To Review, 1 To Rework, 3 Todo”). NO analysis — proceed immediately.

Phase 3: Context Review (before Todo tasks)

Before delegating a Todo task, verify its plan against current codebase:

1. Load task description (get_issue or Read task file)
2. Extract referenced files from task plan
3. Check: Do files exist? Have related files changed? Are patterns still valid?
4. Decision: No conflicts → proceed | Minor changes → update task, proceed | Major conflicts → ask user

Skip Context Review for: To Review tasks, To Rework tasks, test tasks when impl freshly Done, tasks created <24h ago.

Phase 4: Task Loop

For each task by priority (To Review > To Rework > Todo):

1. Delegate to worker via Task tool (see Worker Invocation table)
2. Reload metadata after subagent completes (task count may increase — ln-402 creates [BUG] tasks)
3. If worker sets status != To Review → STOP and report
4. After ln-401/ln-403/ln-404: Immediately invoke ln-402 on same task

Execute → Review → Next. Never skip review. Never batch reviews.

Phase 5: Completion

When all tasks Done:

1. Set Story status to To Review (Linear: update_issue(id, state: "To Review"); File: Edit the **Status:** line)
2. Update kanban: move Story to To Review section
3. Report final status with task counts

• ⚠️ NEVER set Story to Done. Only the quality gate (5XX) can mark Story as Done after full quality check.
• Recommended next step: quality gate for code quality and regression checks

Worker Invocation

CRITICAL: Executors (ln-401/ln-403/ln-404) use Task tool for context isolation. Reviewer (ln-402) runs inline via Skill tool in main flow.

Executors (ln-401/ln-403/ln-404) — Task tool (isolated context):

Task(description: "[Action] task {ID}",
     prompt: "Execute {skill-name} for task {ID}. Read skill from {skill-name}/SKILL.md.",
     subagent_type: "general-purpose")

Reviewer (ln-402) — Skill tool (main flow):

Skill(skill: "ln-402-task-reviewer", args: "{task-ID}")

Formats

TodoWrite (mandatory)

Before each task, add BOTH steps:

1. Execute [Task-ID]: [Title] — mark in_progress when starting
2. Review [Task-ID]: [Title] — mark in_progress after executor, completed after ln-402

Critical Rules

1. Branch isolation: All work in feature/{story-id}-{slug}. Never commit to main/master
2. Metadata first: Never load task descriptions in Phase 2; workers load full text
3. One task at a time: Pick → delegate → review → next. No bulk operations
4. Only ln-402 sets Done: Stop and report if any worker leaves task Done or In Progress
5. Source of truth: Trust Linear metadata (Linear Mode) or task files (File Mode)
6. Story status: ln-400 handles Todo→In Progress→To Review. NEVER set Story to Done — only the quality gate (5XX) can do that after full quality check
7. Commit policy: Only ln-402 commits code. Workers (ln-401/ln-403/ln-404) leave changes uncommitted for ln-402 to review and commit with task ID reference.
8. [BUG] tasks: ln-402 may create new [BUG] tasks mid-review. After metadata reload, reprioritize — new tasks processed in next loop iteration.

Anti-Patterns

• ❌ Running mypy/ruff/pytest directly instead of skill invocation
• ❌ “Minimal quality check” then asking “Want me to run full skill?”
• ❌ Skipping/batching reviews
• ❌ Self-setting Done status without ln-402
• ❌ Executors bypassing Task tool subagent (ln-402 is exception — runs inline)

ZERO TOLERANCE: If running commands directly instead of invoking skills, STOP and correct.

Plan Mode Support

When invoked in Plan Mode (agent cannot execute), generate execution plan instead:

1. Build task execution sequence by priority
2. For each task show: ID, Title, Status, Worker, expected status after
3. Write plan to plan file, call ExitPlanMode

Plan Output Format:

## Execution Plan for Story {STORY-ID}: {Title}

| # | Task ID | Title | Status | Executor | Reviewer |
|---|---------|-------|--------|----------|----------|
| 1 | {ID} | {Title} | {Status} | ln-40X | ln-402 |

### Sequence
1. [Execute] {Task-1} via ln-401-task-executor
2. [Review] {Task-1} via ln-402-task-reviewer
...

Definition of Done

• Working in correct feature branch (verified in Phase 1)
• Story and task metadata loaded; counts shown
• Context Review performed for Todo tasks (or skipped with justification)
• Loop executed: all tasks delegated with immediate review after each
• Story set to To Review (NOT Done); kanban updated
• Final report with task counts and recommended next step: quality gate

Reference Files

• Orchestrator lifecycle: shared/references/orchestrator_pattern.md
• Task delegation pattern: shared/references/task_delegation_pattern.md
• Auto-discovery patterns: shared/references/auto_discovery_pattern.md
• Plan mode behavior: shared/references/plan_mode_pattern.md
• Storage mode detection: shared/references/storage_mode_detection.md
• Executors: ../ln-401-task-executor/SKILL.md, ../ln-403-task-rework/SKILL.md, ../ln-404-test-executor/SKILL.md
• Reviewer: ../ln-402-task-reviewer/SKILL.md
• Auto-discovery: CLAUDE.md, docs/tasks/kanban_board.md

Version: 4.0.0 Last Updated: 2026-01-29