Dispatch

Dispatch Workers: $ARGUMENTS

You are an orchestrator dispatching workers for tasks. By default, each worker runs in its own git worktree for true filesystem isolation. With --sequential, workers execute directly on the current branch without worktrees.

Parse Arguments

Arguments: $ARGUMENTS

Parse the following patterns:

• --count N — Auto-select N ready tasks
• --sequential — Execute tasks one at a time on the current branch (no worktree isolation, no PR per task). Use for dependent/sequential tasks that need to see each other's changes.
• --plan-first — Force all workers into plan approval mode
• --no-plan — Disable auto risk detection, all use bypassPermissions
• --yes — Skip dispatch confirmation (used by /auto-run for autonomous operation)
• --model opus|sonnet — Override model selection for all tasks
• <task-id> — Specific task to dispatch
• <task-id>:"context" — Task with custom context (e.g., INT-15:"Use PriceCache")

If no arguments provided, default to --count 3.

Step 1: Identify Tasks

If --count N was specified (or defaulted):

Query ready tasks: call list_issues(state=Todo) to get candidate tasks. For each, call get_issue(id, includeRelations=true) and filter to those with empty blockedBy arrays. These are the "ready" tasks.

Select the first N tasks that are:

• High priority first — Urgent > High > Normal > Low
• Not already In Progress — Only Todo/Backlog tasks with no blockers

If specific tasks were provided:

Validate each task exists by calling get_issue(id=<task-id>, includeRelations=true).

Step 2: Worktree Cleanup

If --sequential was specified, skip this step (no worktrees used).

Prune any orphaned worktrees before spawning new workers:

git worktree prune
git worktree list

Step 2.1: Validate Sequential Prerequisites

Only if --sequential was specified.

CURRENT_BRANCH=$(git branch --show-current)
if [ "$CURRENT_BRANCH" = "main" ] || [ "$CURRENT_BRANCH" = "master" ]; then
  echo "ERROR: Cannot use --sequential on main/master. Switch to a working branch first."
  # STOP — do not proceed
fi
echo "Sequential mode: tasks will execute directly on branch $CURRENT_BRANCH"

If on main/master, abort dispatch with a clear error. Sequential mode commits directly to the current branch — never to main.

Step 2.5: Risk Assessment (Plan Mode Detection + Risk Tiers)

Skip if --no-plan was specified. If --plan-first was specified, mark ALL tasks [PLAN].

Risk Tier Configuration

First, check for a project-level review config:

cat .claude/review.json 2>/dev/null || cat .claude/risk-tiers.json 2>/dev/null || echo "No review config found"

With review config: Match task file paths (from issue description or related files) against tier patterns. The highest matching tier determines the task's risk level:

Risk Tier	Dispatch Mode
critical	[PLAN]
high	[PLAN]
medium	[AUTO]
low	[AUTO]

Without review config (keyword fallback): For each task, check its title + description (from get_issue) for high-risk keywords (case-insensitive):

• Security: auth, authentication, authorization, encrypt, secret, password, token, credential
• Data: migration, migrate, schema change, drop table, delete data
• Financial: payment, billing, subscription, transaction
• Architectural: architecture, redesign, rewrite, refactor core

Mark matching tasks [PLAN], others [AUTO].

Step 2.7: Model Selection

If --model flag was specified: Use that model for ALL tasks (overrides all other logic).

With review config:

Risk Tier	Model
critical	opus
high	opus
medium	sonnet
low	sonnet

Without review config (keyword fallback):

• Keywords architecture, security, auth, migration, rewrite, redesign in task title/description → opus
• Everything else → sonnet

Display the model in the dispatch summary: [AUTO/sonnet] or [PLAN/opus].

Step 3: Generate Context

For each task without explicit context:

1. Read the task details from get_issue(id=<task-id>, includeRelations=true)
2. Check what other tasks are currently in progress for situational awareness by calling list_issues(state="In Progress")
3. Look for related patterns in the codebase:
   • Check if similar files exist that the worker should follow
   • Look for recently completed dependencies
4. Generate a brief (1-2 sentence) context that helps the worker start faster

Example contexts:

• "Use sector_etfs.txt format from existing tickers/ directory"
• "Follow the pattern in backtesting/cache.py for data storage"
• "Depends on completed INT-14, can use its output"

Step 4: Confirm Dispatch

Present a summary to the user:

If --sequential:

Ready to dispatch N tasks SEQUENTIALLY on branch <current-branch>:
(Tasks execute one at a time, directly on this branch — no worktrees, no PRs per task)

1. <task-id> (P1 <type>) [SEQ/opus]: <title>
   Context: "<generated or provided context>"

2. <task-id> (P2 <type>) [SEQ/sonnet]: <title>
   Context: "<generated or provided context>"

...

If parallel (default):

Ready to dispatch N workers (worktree-isolated):

1. <task-id> (P1 <type>) [PLAN/opus]: <title>
   Context: "<generated or provided context>"

2. <task-id> (P2 <type>) [AUTO/sonnet]: <title>
   Context: "<generated or provided context>"

...

The [PLAN/AUTO/SEQ] tags indicate dispatch mode. The [opus/sonnet] tags indicate the model selection from Step 2.7.

If --yes was specified: Skip the AskUserQuestion confirmation and proceed directly to Step 5.

Otherwise, use AskUserQuestion to confirm:

Ask: "Confirm dispatch of N workers?"

• Options: "Yes, dispatch" / "No, cancel"
• multiSelect: false

Wait for explicit user confirmation before proceeding.

If user selects "No, cancel", abort dispatch.

Step 5: Spawn Workers

If --sequential: use the Sequential Spawn path below. Otherwise: use the Parallel Spawn path (default).

---

Parallel Spawn (default)

Spawn each worker as a subagent with worktree isolation using the Agent tool. Launch all workers in a single message with multiple Agent tool calls for maximum parallelism.

For each task, use the Agent tool with:

• isolation: "worktree" — Each worker gets its own git worktree
• run_in_background: true — Workers run concurrently
• mode: "bypassPermissions" for [AUTO] tasks, mode: "plan" for [PLAN] tasks
• model: "<selected>" from Step 2.7
• name: "<task-id>" — Addressable by task ID

Spawn prompt for [AUTO] tasks:

You are an autonomous worker in your own isolated git worktree. Your task:

<task title and description from get_issue>

Context: <generated context>

Currently in progress (other workers): <list from list_issues state=In Progress, if any>

Instructions:
1. Run `/start-task <task-id>` to claim the task and verify your environment
2. Implement the task according to the acceptance criteria
3. Run `/finish-task <task-id>` when tests pass and implementation is complete

CRITICAL CONTRACT:
- You MUST run /finish-task before completing. A task without a session summary is invisible to coordination.
- If you cannot complete the task, write a partial session summary explaining why before exiting.
- Your worktree will be cleaned up automatically after you finish.

Spawn prompt for [PLAN] tasks:

You are an autonomous worker in your own isolated git worktree, spawned in PLAN MODE. Your task:

<task title and description from get_issue>

Context: <generated context>

Currently in progress (other workers): <list from list_issues state=In Progress, if any>

Instructions:
1. Run `/start-task <task-id>` to claim the task and gather context
2. Create a detailed implementation plan
3. Your plan will be reviewed before you can proceed to implementation
4. After approval, implement the task
5. Run `/finish-task <task-id>` when tests pass and implementation is complete

CRITICAL CONTRACT:
- You MUST run /finish-task before completing. A task without a session summary is invisible to coordination.
- If you cannot complete the task, write a partial session summary explaining why before exiting.
- Your worktree will be cleaned up automatically after you finish.

---

Sequential Spawn (--sequential)

Execute tasks one at a time, in order, directly on the current branch. Each worker sees the previous worker's commits.

For each task in order, use the Agent tool with:

• NO isolation parameter — worker shares the orchestrator's filesystem
• run_in_background: false — orchestrator waits for completion before starting the next task
• mode: "bypassPermissions" (sequential tasks are pre-approved by dispatch confirmation)
• model: "<selected>" from Step 2.7
• name: "<task-id>" — Addressable by task ID

Sequential spawn prompt:

EXECUTION_MODE: sequential
BRANCH: <current branch name>

You are an autonomous worker executing DIRECTLY on branch <current-branch>. No worktree isolation — your changes are immediately visible to the orchestrator and subsequent workers.

Your task:

<task title and description from get_issue>

Context: <generated context>

Instructions:
1. Run `/start-task <task-id>` to claim the task and verify your environment
   - You are already on the correct branch. Do NOT create a new task branch.
2. Implement the task according to the acceptance criteria
3. Run `/finish-task <task-id> --direct` when tests pass and implementation is complete
   - The --direct flag skips PR creation — your commits go directly to this branch.

CRITICAL CONTRACT:
- You MUST run /finish-task with the --direct flag before completing.
- If you encounter merge conflicts or cannot complete the task, STOP immediately and report why. Do NOT retry merge strategies.
- Your commits accumulate on this branch for later review via /milestone-review.

After each sequential worker completes:

1. Verify the worker exited successfully (check return value)
2. If the worker failed: STOP the chain. Report which task failed and which tasks remain. Do not dispatch the next task.
3. If the worker succeeded: proceed to the next task in the list.

If all sequential tasks complete successfully: Proceed to Step 6.

---

Step 6: Post-Dispatch Summary

If --sequential:

Sequential dispatch complete: N tasks executed on branch <current-branch>

Tasks completed:
1. <task-id> [SEQ/sonnet]: <title> ✓
2. <task-id> [SEQ/sonnet]: <title> ✓
...

All commits are on branch <current-branch>.
No PRs created — review accumulated changes with /milestone-review.

IMPORTANT: Before ending this session, run /reconcile-summary to sync
all worker results with the task board.

If parallel (default):

After all workers are spawned, provide a summary:

Dispatch complete: N workers spawned (worktree-isolated)

Workers:
1. <task-id> [AUTO/sonnet]: <title>
2. <task-id> [PLAN/opus]: <title>
...

Each worker has its own git worktree — no filesystem conflicts possible.

Workers will:
1. Run /start-task to claim the task
2. Implement the task
3. Run /finish-task when tests pass (creates PR, session summary)

Workers are running in the background. You will be notified as each completes.
After all workers finish, run /reconcile-summary to process their results.

IMPORTANT: Before ending this session, run /reconcile-summary to sync
all worker results with the task board.

Error Handling

• No ready tasks: Suggest running /orient first to identify work
• Task doesn't exist: Skip it, warn the user, continue with valid tasks
• All tasks invalid: Abort with clear error message
• Worker spawn fails: Report the error, continue with remaining tasks
• Sequential on main branch: Abort with clear error — must be on a working branch
• Sequential mid-chain failure: Stop chain, report failed task and remaining tasks

Examples

Auto-select 3 tasks (default — parallel, worktree-isolated):

/dispatch

Auto-select specific count:

/dispatch --count 5

Specific tasks:

/dispatch INT-15 INT-16

With custom context:

/dispatch INT-15:"Use existing ticker format"

Force all tasks to use Opus:

/dispatch --count 3 --model opus

Sequential execution (tasks run one at a time on current branch):

/dispatch --sequential INT-14 INT-15 INT-16

Sequential with auto-run (autonomous):

/dispatch --sequential --count 3 --yes