The library

wf-swarm-score · Verifying adversarial score spreads and detecting hallucinating ants via 3-role consensus.

research · Exploratory research with parallel hypothesis agents and cross-check synthesis.

ix-adversarial-llm-panel · Complex workflows requiring parallel agents, synthesized judgment, or multi-phase triage.

measure-performance · Perf fixes where file groups are disjoint and can edit in parallel without race conditions.

espeak-audit-all-langs · Finding subtle bugs or policy violations across large codebases that require consistent multi-agent consensus.

espeak-audit-tts-vs-human · Finding subtle bugs or policy violations across large codebases that require consistent multi-agent consensus.

investigate · Orchestrate multi-agent investigate across parallel branches and synthesis.

detect-platforms · workflow automation in specialized problem domain; check artifact name and description.

fieldatlas-deepread · Literature review requiring faithful extraction with verbatim evidence from long academic papers.

enrich-task-tags · Batch tagging homogeneous content (e.g., all AL benchmark tasks) against controlled vocabulary.

vocab-extract · Best for automating complex multi-phase vocab-extract processes at scale.

compound-extract · Mining reusable patterns from code changes while deduplicating existing knowledge.

model-arena-daily · Benchmarking multi-tier LLM responses against a canonical prompt with regression detection and cost-efficiency scoring.

fieldatlas-synth · Grounding research idea generation in corpus citations with novelty-checking and feasibility critique.

model-arena-daily · Daily intelligence on which Claude tier is best for each task type. Cost-disciplined by design: 3 generators + 1 judge =

grounding-audit · Parallel read-only audits against design claims with completeness criticism.

research-pulse-daily · Daily lightweight research pulse: 3 parallel domain scans + synthesis.

driver-menu-build · Interactive menu-driven workflows with dynamic branching.

multi-modal-sweep · Orchestrate multi-phase multi-modal-sweep tasks with structured verification and recovery.

indicator-why-believed · Orchestrate multi-phase indicator-why-believed tasks with structured verification and recovery.

paper-review-fanout · Parallel review and triage workflows across multiple criteria or models.

fieldatlas-validate · Environment or artifact verification with mechanical pass/fail detection against ground truth.

screenspot-param-sweep · Best for adversarial evaluation across agent personas.

classify-improve · improving classifiers via data-driven signal mining from real failures

scout-and-combine · fan-out parallel research with synthesized results

rebench-binary-survey · Comprehensive multi-phase scan across heterogeneous source corpus with parallel validation.

model-convert-benchmark · Multi-phase model-convert-benchmark orchestration with structured phase control and receipts.

references-audit-classify · Comprehensive multi-phase scan across heterogeneous source corpus with parallel validation.

cf-parallel-harvester-rawload · Append-loading UK Contracts Finder historical data (2016-2026) into BigQuery in resumable 2-month shards.

news-graph-extract · Building knowledge graphs from news corpora where entities and relationships must be structured for Neo4j ingestion.

find-kitchen33-datasets · When you need systematic discovery and verification of specialized datasets across multiple sources.

alberta-retention-tournament-verify · When you need multi-phase orchestration with parallel agents across a complex workflow.

driver-menu-build · Building curated driver catalogs from event sources with blind naming.

wf-analyze · Go/no-go decisions on architecture changes when you want scored evidence from multiple dimensions plus adversarial challenge built in.

research · Synthesizing evidence on academic/technical topics when you need multi-modality (text+code+data) sweep plus two-stage extraction-enrichment pipeline.

{PROJECT_NAME}-autoresearch · Use when workflow operations need {PROJECT_NAME} autoresearch.

paper-review-literature · Use when workflow operations need paper review literature.

review-paper · Use when workflow operations need review paper.

deep-research · Use when workflow operations need deep research.

step3-specificity-methodology · Auditing bispecific therapeutic window via parallel research, synthesis, and adversarial critique.

ingest-mcluhan-understanding-media · Ingesting large multi-chapter books with deduplication and cross-chunk synthesis via chunked planners.

deep-research · Running multi-source research with web fetching, fact verification, and cited synthesis.

research · Fan-out web research, dedup sources, synthesize with citations, and fact-check claims.

verify-findings · Verify code audit findings using 3 independent skeptic lenses.

research-pipeline · Execute research through fan-out collection, dedup, and analysis phases.

claim_verifier · Verify document claims are grounded in cited sources.

cp-lit-scan · Systematically surveying a research frontier across multiple sub-topics to identify novel research gaps.

research-digest · Aggregating research signals from 7 heterogeneous sources with real-time relevance ranking and dedup.

mhpp-10-ablation · Ablation studies where solver agents invoke external tools themselves (agentic-tool pattern, not pre-generation).

contracts-finder-backfill · Large-scale procurement time-series ingestion where idempotency and resumability are critical.