← ClaudeAtlas

shared-brief-parallel-fanoutlisted

Use when a task splits into N same-shaped units (audit a corpus by cluster, author N notebooks, tag N subjects, scan N transcripts) and you fan out parallel agents from ONE shared brief — the brief artifact, uniform output contract, agent-vs-script split, and central integration.
hjr15/claude-kit · ★ 0 · AI & Automation · score 73
Install: claude install-skill hjr15/claude-kit
# Shared-Brief Parallel Fanout ## Overview dispatching parallel agents (superpowers) gives the generic rule: one agent per independent domain, dispatch concurrently, integrate. This skill is the recipe for the *homogeneous* variant — N units of the **same shape** (one per cluster / notebook / subject / transcript) — where the leverage comes from a **single shared brief** plus a **uniform output contract** so the parent can mechanically integrate. **Core principle:** write the brief once, make every agent read it in full, make every agent emit the same structure, then integrate centrally. Agents do prose judgment; scripts do mechanical or fragile-format edits. ## When to Use - A task partitions into N comparable units that don't share files (corpus clusters, per-subject pages, per-strand notebooks, per-session transcripts). - Each unit is a bounded "read the brief → do the work → report in the agreed shape" task. - You want consistency across the N outputs (same columns, same template, same dedup baseline). Not this skill: heterogeneous independent failures (use dispatching parallel agents directly); N tasks that each need their own isolated git branch (use parallel worktree tasks); kicking off N *sister sessions* via handoff briefs (use parallel handoff brief fanout). ## The recipe ### 0. Preflight before fanning out For content-authoring fanouts, don't dispatch N agents against an unproven template. First author the hub + **one** fully-finished sample unit yourself,