create-task

/create-task — Create a New Task

Read .claude/project-config.json. Extract monday.productId — the Products-board item ID this task belongs to. Missing → STOP and tell user to set it.

Workflow

Phase 0: Investigate (smart default — see /dev-tasks:investigate-request)

Before any createTask write, run /dev-tasks:investigate-request --mode=dedup with the raw input. This finds related existing tasks (active + recently-shipped), in-flight agents on the same scope, recent merges that may have shipped the work, and surfaces ambiguity as BLOCKING/OPTIONAL questions.

Skip case (deterministic): if the caller arg explicitly cites a Monday task/PR/retro ID (regex /#\d{7,10}\b|PR\s*#\d+|retro\s*#\d+/i), skip — the dedup work is done. Otherwise the invocation is the default.

Handle the recommendation:

• SKIP (duplicate / already covered) → abort. Surface the report to the user. Do NOT call createTask. Post a comment on the existing task if context-additive.
• REFINE #N → redirect: invoke /dev-tasks:refine-task <N> with the report's findings. Do NOT call createTask.
• ROUTE → Bug | Feedback | Retro → wrong surface; redirect: createBug / createFeedback / /dev-tasks:file-retro per the report's suggestion. Do NOT call createTask.
• DECLINE (request superseded) → confirm with AskUserQuestion, then act.
• NEW task → proceed to step 1+ using the report's "Recommendation" scope as the default (priority, epic, subtask shape).

BLOCKING questions in the report MUST be resolved via AskUserQuestion before any createTask call. OPTIONAL questions are mentioned in the proceed-message but don't gate action. See the skill for the BLOCKING vs OPTIONAL heuristic.

The legacy cross-surface check (this should be a bug / retro / feedback, not a task) is subsumed by the investigation — the report's "Recommendation" line says so explicitly when relevant.

Standard workflow

1. Cross-surface fallback (only when Phase 0 was skipped via the explicit-ID skip case): would this produce wrong output for a real user? → createBug instead. Workflow/tooling friction? → /dev-tasks:file-retro. Stakeholder feedback? → createFeedback + let /triage-feedback route.

2. Epic assignment (MANDATORY — no orphaned tasks):

   • listEpics(productId: $productId). Auto-match by keyword. If confident, use; else ask user.
   • Bugs/hotfixes: pick product's Maintenance epic (name contains "maintenance" case-insensitive).

3. Type & priority:

   • type: Feature / Fix / Improvement / To Do
   • priority: Critical / High / Medium / Low. Do NOT leave Missing unless genuinely uninferable — Ready to Start gate rejects it.

4. Description + acceptance criteria (both REQUIRED for the gate):

   • description: WHAT and WHY (1–3 paragraphs)
   • acceptanceCriteria: bullet list of definition-of-done items an external party can verify

5. Plan subtasks (REQUIRED — at least one, fully refined):

   • Each subtask: name + description + type + estimatedHours
   • Types: Backend / Test / Documentation / UX-UI / Database / To Do (see task-lifecycle.md)
   • NEVER create a human-testing subtask. Human verification belongs to parent's Waiting for UAT + UAT doc on column doc_mm3adfdg (auto-generated by /ship-pr Phase 4.5).

6. Optional dependency: dependencyIds: [<task-id>, ...] (column dependency_mm0pwbxn). claimTask refuses to start until these are Done.

7. Create: createTask with:

   • status: "Ready to Start" — MCP rejects if any prereq missing; treat rejection as checklist.
   • agentId: "Claude Code CLI" (or creating agent)
   • owner: <whoami>
   • epicId, type, priority, description, acceptanceCriteria
   • subitems: typed+estimated list
   • sprintId: <active> if claimTask follows (else /pickup-task Phase 7 auto-assigns)
   • dependencyIds if step 6 produced any

8. Output: task ID, epic name, final status (should be Ready to Start if gate passed), Monday link.

Arguments

• Description of the task (natural language)

Ready to Start gate (server-side)

MCP rejects unless ALL present:

• type (not "Not Set")
• priority (not "Missing")
• epicId
• description
• acceptanceCriteria
• ≥1 subtask with name + description + type + estimatedHours

createTask runs this gate at creation when status: "Ready to Start". Failed gate → task created at Needs Refinement; missing fields in response. Populate via updateTask / manageSubtasks, then flip status.

In Progress gate

claimTask refuses unless:

• Status Ready to Start
• Task in active sprint (/pickup-task Phase 7 auto-assigns)
• All dependencyIds are Done
• No other agent currently owns

Status flow

1. Ready to Start → /pickup-task claims → In Progress
2. In Progress → all subtasks Done + /ship-pr generates UAT doc → Waiting for UAT
3. Waiting for UAT → human UAT signoff → Pending Deploy to Prod
4. Pending Deploy to Prod → /release-version → Done

See CLAUDE.md "Task Management Lifecycle" for the full table.