Loading...
Loading...
Found 10 Skills
Improve context efficiency through context budgeting, observation masking, prefix or KV-cache strategy, partitioning, token-cost reduction, retrieval scoping, and extending effective context capacity without lowering answer quality. Use when token costs or context budgets constrain a task, tool outputs are verbose, cache hit rate is low, or context must be partitioned across agents.
This skill should be used for improving context efficiency: context budgeting, observation masking, prefix or KV-cache strategy, partitioning, token-cost reduction, retrieval scoping, and extending effective context capacity without lowering answer quality.
A/B test agent variants measuring quality and total session token cost across simple and complex benchmarks. Use when creating compact agent versions, validating agent changes, comparing internal vs external agents, or deciding between variants for production. Use for "compare agents", "A/B test", "benchmark agents", or "test agent efficiency". Do NOT use for evaluating single agents, testing skills, or optimizing prompts without variant comparison.
This skill should be used when the user asks to "optimize context", "reduce token costs", "improve context efficiency", "implement KV-cache optimization", "partition context", or mentions context limits, observation masking, context budgeting, or extending effective context capacity. A core context engineering skill — also activates when the user mentions "context engineering" or "context-engineering" in the context of maximizing information density within token constraints.
Analyzes observability signals from customer GenAI applications with DQL. Reads OpenTelemetry GenAI spans and LLM evaluation bizevents. Use for: golden signals (traffic, errors, latency, saturation); LLM signals (model, provider, tokens); cost/token analytics, usage attribution, and prompt caching; agent signals (tool calls, steps, failures, loop detection, Smartscape topology); conversation/session analytics; guardrails (blocked/truncated responses); and evaluation signals (quality, pass/fail). Trigger: "LLM latency", "token usage by model", "cost by model and provider", "cost per conversation", "who is driving token spend", "do I have prompt caching", "failing agent tool calls", "find runaway agents", "responses truncated or blocked", "failed evaluations", "am I hitting rate limits", "token throughput / TPM", "provider throttling or 429s". Do NOT use for: Davis CoPilot/MCP telemetry (dt-platform), generic service metrics (dt-obs-services), logs (dt-obs-logs), or non-GenAI tracing (dt-obs-tracing).
Track per-agent token usage and flag waste in parallel dispatch. Use after running parallel agents to evaluate cost vs value.
ABSOLUTE MUST to debug and inspect LLM/AI agent traces using PostHog's MCP tools. Use when the user pastes a trace URL (e.g. /llm-observability/traces/<id>), asks to debug a trace, figure out what went wrong, check if an agent used a tool correctly, verify context/files were surfaced, inspect subagent behavior, investigate LLM decisions, or analyze token usage and costs.
Comprehensive context management strategies for cost optimization and infinite-length conversations. Covers server-side clearing (tool results, thinking blocks), client-side SDK compaction (automatic summarization), and memory tool integration. Use when managing long conversations, optimizing token costs, preventing context overflow, or enabling continuous agentic workflows.
Install a pre-configured Claude Code statusline with model, tokens, cost, lines changed, and context usage. A batteries-included alternative to /statusline with formatting and color thresholds already defined.
Benchmark throughput and cost-per-token of a model served on vLLM in Chalk's container infrastructure. Use when someone asks for $/token, $/request, tokens/sec, "how much would it cost to self-host this model", GPU sizing for an LLM, or wants to compare a self-hosted model's economics against a hosted API.