Eval Swarm — Productivity skill for Claude Code
Fan every eval- evaluator agent across slices of the monorepo as a parallel swarm via the eval-swarm workflow — review, adversarially verify P0/P1, then route confirmed defects to break-fix for group.
How to install Eval Swarm
Installs to ~/.claude/skills/macanderson-agent-harness-eval-swarm/SKILL.md
mkdir -p ~/.claude/skills/macanderson-agent-harness-eval-swarm && curl -fsSL https://raw.githubusercontent.com/macanderson/agent-harness/HEAD/commands/eval-swarm.md -o ~/.claude/skills/macanderson-agent-harness-eval-swarm/SKILL.md Restart Claude Code, or start a new session, for it to be picked up.
What Eval Swarm does
description: Fan every eval-* evaluator agent across slices of the monorepo as a parallel swarm via the eval-swarm workflow — review, adversarially verify P0/P1, then route confirmed defects to break-fix for grouped remediation. Optionally scope to a path with an argument.
/eval-swarm
Run the full `eval-*` evaluator fleet across the monorepo as a workflow-orchestrated swarm. Each slice of the tree is reviewed in parallel by every applicable lens (`eval-database`, `eval-typescript`, `
Alternatives in Productivity
- Epic Decompose — Break an epic into task children without creating task branches 243.5k ★
- Worktree Status Check — Verify the current worktree environment and show task details 23.4k ★
- Swarm Worker — You are a swarm worker agent for the {{tentacleName}} tentacle 1.4k ★
Full documentation available on GitHub
View Source RepositoryRelated Skills
Cheap Delegate
Smart cheapest-adequate delegation — route a task to Ollama workers, OpenCode agents, Codex, Sonnet or Opus vi
M Bug Fix
Bug Fix Workflow: Analyze, Reproduce, Identify, Fix, Verify
Bug Mode Arken
Create developer-ready Arken bug stories with complete reproduction context, impact assessment, and fix verifi
Writ Approve
Advance the current Writ workflow phase with an explicit tool-confirmed advance. NOT a replacement for pattern
Claude Technique Evaluator
Evaluate new Claude prompting patterns, tools, or workflow changes. Produces go/no-go recommendations and inte
Musecode
Ultracode" on Muse Spark 1.3 (free) — orchestrate the task as a Workflow where every fan-out agent is the muse
Related Agents
Dr Verifier
Deep-review verify pass. Takes a batch of deduplicated candidate findings and adversarially tries to disprove
Structure Planner
Use after the design review passes to break the work into vertical slices with verification checkpoints. Each
Deal Verifier
Adversarially tries to break proposed contract patches and verifies source-document integrity. Default verdict