Skip to content

rohitg00/pro-workflow / skills/batch-orchestration

batch-orchestration

Unverified

Decompose large-scale changes into independent units and spawn parallel agents in isolated worktrees. Use for migrations, refactors, codemods, and any change touching 10+ files with the same pattern.

Batch Orchestration

The /batch command pattern for large-scale parallel changes.

How It Works

/batch <instruction>
  │
  ├── 1. Research: scan repo, understand scope
  ├── 2. Decompose: split into 5-30 independent units
  ├── 3. Present plan: show units, ask for approval
  ├── 4. Execute: one background agent per unit in isolated worktree
  └── 5. Collect: each agent runs tests and opens a PR

Syntax

/batch Convert all React class components to function components
/batch Add error boundaries to every page component
/batch Migrate from moment.js to dayjs across the codebase
/batch Add OpenTelemetry tracing to all API handlers

The instruction should describe the change pattern, not individual files. The batch system finds the files.

Phase 1: Research

The orchestrator scans the repo to find every instance that matches the instruction:

grep -r "class.*extends.*Component" --include="*.tsx" -l

It builds a complete list of targets and groups them by independence.

Phase 2: Decompose

Each unit must be:

  • Independent — no shared state with other units
  • Self-contained — can be implemented and tested alone
  • Verifiable — has a clear pass/fail criteria

Good units:

Unit 1: Convert src/components/Header.tsx (class → function)
Unit 2: Convert src/components/Footer.tsx (class → function)
Unit 3: Convert src/components/Sidebar.tsx (class → function)

Bad units:

Unit 1: Convert all components in src/components/ (too broad)
Unit 2: Fix issues from Unit 1 (dependent)

Target 5-30 units. Fewer than 5 doesn't justify the overhead. More than 30 and coordination costs grow.

Phase 3: Plan Approval

The orchestrator presents:

BATCH: Convert class components to function components

Found: 18 class components across src/

Units (18):
  1. src/components/Header.tsx — class Header → function
  2. src/components/Footer.tsx — class Footer → function
  ...
  18. src/pages/Settings.tsx — class Settings → function

Per unit: convert class to function, update hooks, run component tests
Estimated: ~2 min per unit, ~5 min total (parallel)

Proceed? (y/n)

Wait for approval. Never spawn agents without explicit confirmation.

Phase 4: Execute

After approval, for each unit:

  1. Create isolated git worktree
  2. Spawn background agent in that worktree
  3. Agent implements the change
  4. Agent runs relevant tests
  5. Agent opens a PR

Agents run in parallel. Each has its own context window and worktree — no conflicts.

[Agent 1] ── worktree-1 ── Header.tsx ── tests pass ── PR #41
[Agent 2] ── worktree-2 ── Footer.tsx ── tests pass ── PR #42
[Agent 3] ── worktree-3 ── Sidebar.tsx ── tests fail ── flagged

Phase 5: Collect

After all agents complete:

  • Summary of pass/fail per unit
  • Links to opened PRs
  • Any units that failed with error details
  • Failed units can be retried individually

Best For

Use CaseWhy Batch Works
API migrationsSame pattern across many endpoints
Dependency upgradesFind/replace + fix across codebase
Codemod-style refactorsMechanical transformation, file by file
Adding instrumentationSame tracing/logging pattern everywhere
Test coverage gapsAdd tests to untested modules independently
Lint rule adoptionApply new rule fixes across all files

Anti-Patterns

Don't BatchWhy
Interdependent changesUnits can't run in parallel if they depend on each other
Shared state modificationsMultiple agents writing to the same config or state file
Architecture changesNeed holistic reasoning, not file-by-file
Schema migrationsDatabase changes must be sequential
Changes requiring human judgment per fileDefeats the purpose of automation

Relationship to Other Patterns

PatternScaleIsolation
Direct edit1-3 filesNone needed
Subagent1 focused taskForked context
Worktree1 feature branchFull repo copy
Agent teams3-5 parallel tasksShared task list
Batch5-30 identical patternFull worktree per unit

Batch is the heaviest tool. Use it when the change is mechanical, repetitive, and the units are truly independent.

Guardrails

  • Always review the decomposition before approving
  • Each agent must run tests before opening a PR
  • Failed units get flagged, not silently skipped
  • Clean up worktrees after all agents complete
  • Review PRs in batches — don't merge blindly
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
discernment-nudgeanthropics/skillsAfter you give a substantive answer or draft that the user may act on — advice or recommendations, drafted artifacts such as goals, plans, pitches, proposals, or emails, estimates or projections, analysis or interpretation of data, factual claims they may rely on, or a multi-step argument — invoke this skill BEFORE finalizing your reply and then, if it applies, append 2-3 short follow-up questions, each tied to something specific in what you just produced, that help the user check key facts, probe the reasoning or assumptions, and notice missing context. Do this at most once per conversation. Skip it when the user asked a trivial how-to or simple lookup, wants a purely educational explanation, asked you only to format, convert, or assemble a file from content they provided, is writing code they will run, is doing creative writing or casual chat, or already asked you to double-check, cite, or review — the skill file explains these boundaries and the exact output format.Stars: 195kUpdated Claude CodeCodexVerifiedInstall
internal-commsanthropics/skillsA set of resources to help me write all kinds of internal communications, using the formats that my company likes to use. Claude should use this skill whenever asked to write some sort of internal communications (status reports, leadership updates, 3P updates, company newsletters, FAQs, incident reports, project updates, etc.).Stars: 195kUpdated Claude CodeCodexVerifiedInstall