Loading...
Loading...
Found 21 Skills
A method for iteratively improving text instructions for agents (skills / slash commands / task prompts / CLAUDE.md sections / code generation prompts) by having unbiased executors run them, then evaluating from both perspectives (executor self-report + instruction-side metrics). Repeat until improvement plateaus. Use immediately after creating or significantly revising a prompt or skill, or when you suspect the reason an agent isn't behaving as expected is due to ambiguity in the instructions.
Use when measuring or improving agent quality and performance — set up evaluators, online monitoring, CI/CD quality gates, observability, or cost optimization. Triggers on: "evaluate my agent", "add evaluator", "measure quality", "quality gate", "run evals", "agent too slow", "why is it slow", "reduce latency", "set up observability", "CloudWatch dashboard", "how much does my agent cost", "cost optimization", "logs not showing up", "logs missing", "spans not found", "eval failing", "eval error", "dev traces", "local traces", "agentcore dev traces", "traces to CloudWatch". Not for debugging errors or crashes — use agents-debug. Slow but correct routes here; broken routes to debug.
Set up and improve harness engineering (AGENTS.md, docs/, lint rules, eval systems, project-level prompt engineering) for AI-agent-friendly codebases. Triggers on: new/empty project setup for AI agents, AGENTS.md or CLAUDE.md creation, harness engineering questions, making agents work better on a codebase. ALSO triggers when users are frustrated or complaining about agent quality — e.g. 'the agent keeps ignoring conventions', 'it never follows instructions', 'why does it keep doing X', 'the agent is broken' — because poor agent output almost always signals harness gaps, not model problems. Covers: context engineering, architectural constraints, multi-agent coordination, evaluation, long-running agent harness, and diagnosis of agent quality issues.
Benchmark any agent skill to measure whether it actually improves performance. Use when the user wants to evaluate, test, or compare a skill against baseline, or when they mention "benchmark", "eval", "skill performance", or "does this skill help". Runs isolated eval sessions with and without the skill, grades outputs via layered grading (deterministic checks + LLM-as-judge), analyzes behavioral signals, and generates a comparison report with a USE / DON'T USE verdict.
Head-to-head comparison of coding agents (Claude Code, Aider, Codex, etc.) on custom tasks with pass rate, cost, time, and consistency metrics
Create code-based evaluators for LangSmith-traced agents with step-by-step collaborative guidance through inspection, evaluation logic, and testing.
Take your AI agent to the next level with full LangWatch integration. Adds tracing, prompt versioning, evaluation experiments, and simulation tests in one go. Use when the user wants comprehensive observability, testing, and prompt management for their agent.
AI Agent learning roadmap and curated resources for building production-ready agents with modern patterns like Claude Code, OpenClaw, skills, MCP, and evaluation
Create a new Harbor task for evaluating agents. Use when the user wants to scaffold, build, or design a new task, benchmark problem, or eval. Guides through instruction writing, environment setup, verifier design (pytest vs Reward Kit vs custom), and solution scripting.