The library

boardroom-blind-review · High-stakes decisions where reviewer independence must be enforced structurally via separate agent contexts.

qa · Budget-gated QA where lite testing is the gate before heavier runs.

parallel-swarm.workflow · Batch operations where each item can be independently processed in parallel.

wait-for-pr-review · CI pipelines where Claude-generated reviews must complete before merging.

sprite-restyle-judge · Game asset QA where visual consistency to canon must be both scored and gated per character.

stale-ticket-drafts · Automated ticket triage where stale tickets need human-readable status drafts without direct edits.

judge-independent-audit · Auditing complex ML systems.

team-refine · Use when implementing team-refine patterns.

review · Use when implementing review patterns.

phase-b0-cyclebreak · Use when implementing phase-b0-cyclebreak patterns.

build-viz · Use when implementing build-viz patterns.

emergence-scan · Use when implementing emergence-scan patterns.

t · Use when implementing t patterns.

qa-parallel-suite · Budget-gated QA where lite testing is the gate before heavier runs.

bench-s0-monolith · Use when implementing bench-s0-monolith patterns.

trace-concurrency-cap · empirically validating the workflow runtime's true parallel capacity before launching production workloads

phase-b2-fix-bugs · applying pre-verified bug fixes across multiple crates without sequencing delays

phase-d-bundler-perfile · methodically eliminating cascading compiler errors in a large module via per-file focus

phase-b2-ungate-tier · exposing and fixing draft code dependencies before moving to verification pass

phase-d-bundler-shard · parallelizing fix work on a single large file (bundle_v2) across multiple agents without conflicts

frontier-tail-survey · auditing true build state across multiple frontiers without false confidence from presence of stub files

cr-threshold-gate · mechanical OR-gated CR dispatch (LOC > 50 OR core-module hit) without human judgment

ship-item · automated production merge gate with adversarial reviewer integration and carry-forward tracking

translate-keys · Localizing customer-facing UI (like ordering combos) into 37 languages with strict JSON schema + placeholder + markup preservation.

phase-b0-moveout · systematically breaking dependency cycles before move-in pass via vtable/tag+ptr/hook delegation

phase-b0-movein · resolving forward references established in move-out via symbol materialization in target crates

phase-b2-fill · on-demand symbol provisioning as downstream blockers surface during multi-phase porting

benchmark-audit · spotting performance regressions in CI without re-running expensive GPU benchmarks

emit-beads · Bulk-importing planning data into beads while maintaining idempotency and dependency integrity

phase-b2-fill-blocked · Unblocking Rust port tasks by filling symbol gaps from a structured dependency report

grill · Finding unstated assumptions and failure modes before design is locked

doc-drift-audit · Catching documentation claims that grep/structural validators miss (citations, superlatives, stale enumerations, business facts, capability contradictions).

codemaps-extract · Architectural review insight (coupling risk, latent contracts, layering) layered on top of deterministic go list/go doc baseline for medium-to-large codebases.

vertical-conformance-sweep · Catching recurring plugin-contract bugs (schema strip, action/tool mismatch, missing files, chip wiring) that occur after editing or before release.

impl-build · Converting an approved plan into code while surfacing every deviation and stopping on blocking gaps, preserving the plan intent with minimal changes.

agent-intention-redteam · Finding how an agent behaves unexpectedly (specification-gaming, misalignment, emergent behavior) before production deployment, complementing generic jailbreak scanners.

review-panel · Deep pre-merge gate for high-risk changes (T3: migrations, RLS, coaching prompts, consent flows) where multiple correctness/security/perf/compliance angles must pass.

lead-contacts-find · Scalable concurrent discovery and seeding of <=5 decision-maker contacts per stale company (part of lead enrichment pipeline; 3 in flight concurrency).

sm-proposal-research · Designing high-stakes proposals with ironclad guardrails (firewall, polarity, ADR-18 floor) and domain-agnostic, a11y-compliant mockups that operator must confirm before code.

cleanup-issues-prs · Recurring issue/PR backlog grooming when needs to flag fixed/stale/duplicate tickets and stuck PRs without manual review

bug-hunt · Recurring correctness sweeps where bugs must be proven by test before shipping (not style/design audits)

tune-article · One-off fine-tuning of published techfromalex.com articles; never auto-writes to live site

auditar-confiabilidad · One-time or recurring reliability audit of all harness automations; identifies decision-log gaps and backoff issues

feature-recon · BEFORE building any feature/module; avoids reimplementing existing code or merging dead branches

self-improve-forecaster · Improving a forecasting playbook on a domain-specific dataset without web research (airgapped)

layout-hillclimb · Incremental layout algorithm improvements where each fix is validated against baseline + other fixtures (HYBRID gate)

routing-collision-sweep · Proactive skill routing audit before routing causes user confusion; prevents silent shadowing

docs-accuracy-audit · Keeping distributed documentation synchronized with code changes without manual review

mercury-codebase-audit · Repo-wide sweeps that cannot fit in one conversation, with plausible-but-wrong findings filtered by adversarial skeptics

mcp-discovery-survey · Deterministic monthly/quarterly MCP ecosystem surveys that feed implementation planning without re-discovering