agent-pattern-selectorlisted

# agent-pattern-selector The front door. A read-only triage that maps a one-line problem to the harness skill that owns it — and, just as often, reminds you that you do not need the fancy thing yet. It writes nothing and runs nothing; it points. ## When to use - You know the symptom ("too slow", "it forgets", "needs sign-off") but not the tool. - You are about to build something elaborate and want a sanity check first. - Onboarding to the harness and looking for the right entry point. ## The prime directive > Find the simplest thing that works, and only add complexity when it demonstrably earns its keep. Most "agent" problems are solved by a better prompt, one good tool, or a single workflow shape — not by more agents. Reach for orchestration and multi-agent topologies last, not first. ## Triage table | Your problem | Start here | |--------------|-----------| | "Is the new prompt/model actually better?" / I need a regression gate | [[eval-loop-builder]] | | "Grade this one open-ended answer against a rubric" | [[eval-judge]] | | "Block PII / enforce a format / refuse unsafe output at runtime" | [[guardrail-designer]] | | "What did the agent do? where's the time/cost going?" | [[observability-tracer]] | | "A human must approve before this runs" | [[hitl-gate]] | | "Too slow / too expensive / caching won't hit" | [[cost-latency-optimizer]] | | "How should I structure these steps?" (chain/route/parallel/refine) | [[orchestrator-patterns]] | | "Do I need multiple agents,