From metapowers
Implement a task by coordinating an agent team with parallel prototypes, multi-area code review, fix cycles, PR creation, and comparative evaluation. Use when the user wants to explore multiple implementation approaches for a spike, prototype, or feature.
How this skill is triggered — by the user, by Claude, or both
Slash command
/metapowers:prototype-and-reviewThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Implement the beads issue `$ARGUMENTS` by coordinating an agent team. Create parallel prototypes, review them, fix issues, create PRs, and evaluate.
Implement the beads issue $ARGUMENTS by coordinating an agent team. Create parallel prototypes, review them, fix issues, create PRs, and evaluate.
bd show $ARGUMENTS to understand the task, its dependencies, and deliverables.bd update $ARGUMENTS --status=in_progressCreate a beads subtask for each prototype:
bd create --title="Prototype X: <approach>" --description="<details>" --type=task --priority=1
bd dep add <subtask-id> $ARGUMENTS --type=parent-child
For each prototype, create child subtasks for the implementation steps. Link them with parent-child and blocking dependencies. The agent will update these as it progresses.
Create an evaluation task blocked by all prototype tasks:
bd create --title="Evaluate prototypes and recommend direction" --description="<criteria>" --type=task --priority=1
bd dep add <eval-id> $ARGUMENTS --type=parent-child
bd dep add <eval-id> <prototype-1-id> --type=blocks
bd dep add <eval-id> <prototype-2-id> --type=blocks
Pre-create git worktrees for each prototype (the isolation: "worktree" parameter may not work reliably in all environments):
git worktree add /tmp/<prototype-name> -b <branch-name>
cd /tmp/<prototype-name> && pnpm install
Create a team via TeamCreate.
Launch implementation agents in parallel — one per prototype. Use subagent_type: "general-purpose" with mode: "bypassPermissions" and run_in_background: true. Do NOT use isolation: "worktree".
Each agent's prompt MUST include:
cd /tmp/<worktree-path> as the FIRST instructionVerify worktree isolation after launch — confirm git worktree list shows separate entries and git branch --show-current on main hasn't changed.
Wait for all implementation agents to complete.
Launch review agents in parallel for each prototype. Use subagent_type: "feature-dev:code-reviewer" with run_in_background: true.
Group the 7 review areas into 3 focused agents per prototype:
Area 1 — Coding Style:
any, good generics use)Area 5 — Debugging Cruft:
console.log not part of intentional outputArea 6 — Consistent Formatting:
Area 2 — Efficiency and Algorithm Appropriateness:
Area 4 — Robust Error Handling:
Area 3 — Error/Warning Cleanup:
tsc --noEmit and report errors@ts-ignore, @ts-expect-error, as anyArea 7 — Functional Test Coverage:
Each reviewer rates issues as:
Only MUST FIX items are reported (with file:line citations).
For each prototype:
cd /tmp/<worktree> && git push -u origin <branch>gh pr create including:
Claim the evaluation beads task.
Launch 3 evaluation agents in parallel (subagent_type: "feature-dev:code-explorer"), each analyzing from a different perspective:
Collate evaluations into a recommendation with justification.
Post evaluation summaries as comments on the respective PRs.
git worktree remove /tmp/<name>git status shows clean, up to date with origin.bd) for task tracking — never markdown TODOs or TaskCreateisolation: "worktree"isActive flags in team config if TeamDelete failsnpx claudepluginhub galexy/metapowers --plugin metapowersOrchestrates a full build pipeline from PRD: plans tasks, implements in parallel, and reviews results. Routes simple work to a lighter workflow.
Implements features using parallel subagents with scope control, reflection, and MCP servers for memory/context. Activates on implement/build/create requests in JS/TS projects.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.