Loading...
Loading...
Found 29 Skills
This skill should be used when the user asks to "diagnose context problems", "fix lost-in-middle issues", "debug agent failures", "understand context poisoning", or mentions context degradation, attention patterns, context clash, context confusion, or agent performance degradation. Provides patterns for recognizing and mitigating context failures.
Agentica server + Claude proxy setup - architecture, startup sequence, debugging
Use when diagnosing agent failures, debugging lost-in-middle issues, understanding context poisoning, or asking about "context degradation", "lost in middle", "context poisoning", "attention patterns", "context clash", "agent performance drops"
This skill should be used when inspecting, analyzing, or querying Claude Code session logs. Use when users ask about session history, want to find sessions, analyze context usage, extract tool call patterns, debug agent execution, or understand what happened in previous sessions. Essential for understanding Claude Code's ~/.claude/projects/ structure, JSONL session format, and the erk extraction pipeline.
This skill should be used when the user asks to "diagnose context problems", "fix lost-in-middle issues", "debug agent failures", "understand context poisoning", or mentions context degradation, attention patterns, context clash, context confusion, or agent performance degradation. A core context engineering skill — also activates when the user mentions "context engineering" or "context-engineering" in the context of diagnosing and mitigating context failures.
Debugs errors and traces failures in AI agents and their tools. Use this skill when the user says: "the agent is failing", "tool call not working", "error in the pipeline", "debug this", "why is the agent doing X instead of Y", "trace the execution", "agent is stuck", "infinite loop", "model response won't parse", "context overflow". Identifies context errors, infinite loops, malformed tool calls, response parsing issues and subagent conflicts.
Instruments Python and TypeScript code with MLflow Tracing for observability. Triggers on questions about adding tracing, instrumenting agents/LLM apps, getting started with MLflow tracing, or tracing specific frameworks (LangGraph, LangChain, OpenAI, DSPy, CrewAI, AutoGen). Examples - "How do I add tracing?", "How to instrument my agent?", "How to trace my LangChain app?", "Getting started with MLflow tracing", "Trace my TypeScript app"
Auditing memory traces and debugging.
Debugging toolkit for AI agents. Diagnose symptoms via memory cache -> behavior cache -> codebase search, trace data flow, git-bisect bad commits, and compare output directories.
Retrieve and analyze simulation results from a Coval run. Use when user wants to review evaluation outcomes or debug agent behavior.
Use when your agent or environment is broken — wrong answers, errors, timeouts, tool failures, or CLI issues. Reads traces and logs to diagnose root causes. Also checks prerequisites when the CLI itself isn't working. Triggers on: "agent not working", "wrong answer", "agent error", "tool call failing", "debug agent", "check logs", "read traces", "broken", "500 error", "424 error", "model access denied", "command not found", "stuck in DELETING", "maxVms exceeded", "cold start diagnosis", "cold start slow", "agentcore create error", "create failed", "exit code 7", "connection refused local dev". Not for deploy failures — use agents-deploy. Not for performance tuning without errors — use agents-optimize. Not for VPC configuration — use agents-build. Not for observability setup or missing logs — use agents-optimize.
Drives a multi-turn conversation with a LiveKit agent running locally to see what it does. Use when the user says "test my agent", "try my agent", "does this work", "why did it call that tool", "it says the wrong thing when I ask X", "test this change", or whenever you have edited an agent and need to check how it behaves. Wraps `lk agent debugger`: start the agent in text mode, send turns, read the tool calls, handoffs, errors and logs behind each reply, and restart after an edit. It runs without audio or a LiveKit room at one LLM call per turn, so it is the preferred way for a coding agent to live-test during development, and the default when the user says "test" without naming unit tests or simulations.