clone-sprint

Sprint-scoped MVP builder. Consumes clone-research/{slug}/ output, decomposes into a beads task DAG, and executes one sprint per invocation.

Prerequisites

1. Research output must exist: clone-research/$ARGUMENTS/ with 9 standardized docs
2. Beads CLI installed: bd must be on PATH
3. Git repo initialized: working directory must be a git repository

If research is missing, tell the user to run /clone-research $ARGUMENTS first. If bd is not installed, instruct: npm install -g @beads/bd

Execution Mode

Detect state and execute the appropriate phase:

References: workflow.md

Research Contract

Clone-sprint expects clone-research output in a specific format. If these change, both skills must be updated together.

Docs 01 and 07 provide context but have no strict structural requirements.

Phase 1: Initialize & Decompose

When: First invocation — no .beads/ directory exists.

1. Verify research completeness — read 00-INDEX.md, check all docs exist, abort if doc 05 missing
2. Read all available research docs from clone-research/$ARGUMENTS/
3. Initialize beads: bd init --quiet
4. Create scaffold epic (priority 1, labeled scaffold) with subtasks for project init, database, design system, CI
5. Extract bounded contexts from 05-domain-model.md → create one epic per context
6. Extract P0 features from 02-feature-priority-matrix.md → create subtasks grouped by context
7. Build rich task descriptions embedding entities (doc 05), API endpoints (doc 08), acceptance criteria (doc 03), design tokens (doc 09), suggested file targets (doc 06)
8. Wire dependencies: scaffold blocks all epics; follow entity relationships and user journeys
9. Validate DAG: bd swarm validate — check for cycles and compute parallelism
10. Report DAG summary to user (epic count, task count, estimated sprints)
11. Immediately proceed to Phase 2 with ready scaffold tasks

References: task-decomposition.md, beads-integration.md

Phase 2: Sprint Execution

When: bd ready --json returns tasks.

1. Query ready tasks: bd ready --json
2. Select batch (2-3 tasks):
   • Higher priority first
   • From different bounded contexts (minimize file overlap)
   • Check file targets for overlap — never assign overlapping tasks to same sprint
   • Hard limit: 3 agents maximum
3. Claim tasks: bd update <id> --claim for each selected task
4. Determine agent type per task from labels (fullstack/data/api/ui/infra)
5. Compose agent prompts with common preamble + type-specific instructions + task description
6. Spawn all agents in ONE message using Task tool:
   • subagent_type: "general-purpose"
   • isolation: "worktree"
   • All agents launched in parallel
7. Wait for all agents to complete
8. Merge results serially using bd merge-slot:
   • Order: data → API → fullstack → UI → infra
   • Tier 1: fast-forward (happy path)
   • Tier 2: rebase + retry (another agent merged first)
   • Tier 3: escalate to user (genuine conflict)
9. Close merged tasks: bd close <id> --reason "Merged"
10. Report sprint results:
    • Tasks merged vs failed
    • Overall progress (completed/total, percentage)
    • Ready tasks for next sprint
    • Estimated remaining sprints

References: agent-roles.md, merge-strategy.md, workflow.md

Phase 3: Completion

When: All tasks are closed (bd ready empty, no in_progress tasks).

1. Run bd stats --json for final statistics
2. Present completion report:
   • Total tasks and sprints used
   • Epic-by-epic summary (all checkmarks)
   • Merge conflict count
3. Suggest next steps:
   • Code review
   • Full test suite run
   • Dev server startup
   • P1/P2 feature planning

References: workflow.md

Anti-Patterns

DO NOT decompose by horizontal layer (all migrations → all APIs → all UI). Each task must be a vertical slice within one bounded context.

DO NOT create tasks with thin descriptions. Every task must embed entity definitions, API specs, acceptance criteria, design tokens, and file targets.

DO NOT spawn more than 3 agents per sprint. Merge conflicts scale quadratically.

DO NOT skip bd merge-slot. Direct parallel merges cause race conditions.

DO NOT create more than ~30 tasks for an MVP. If you have more, combine related features into broader slices.

DO NOT assign tasks from the same bounded context to the same sprint unless they have zero file overlap.

Example Invocations

# First run: creates DAG from research + runs scaffold sprint
/clone-sprint linear

# Subsequent runs: executes next sprint
/clone-sprint linear

# After all sprints complete: shows final report
/clone-sprint linear

Typical MVP: 15-20 tasks across 6-8 sprint invocations.