Add cost_estimate and benchmark OpenClaw skills

openclaw/skills/cost_estimate/SKILL.md

@@ -0,0 +1,44 @@
+# cost_estimate
+
+Estimate the cost of running queries through different LLM providers and models. Supports all 12 providers: OpenAI, Claude, Gemini, Mistral, Cohere, Ollama (free), OpenRouter, Groq, xAI, Together, HuggingFace, DeepSeek.
+
+## Parameters
+- `num_queries` (integer, required): Number of queries to project
+- `provider` (string, optional): Specific provider to analyze (default: all)
+- `pipeline` (string, optional): "baseline", "graphrag", or "both" (default: both)
+
+## Returns
+JSON with cost projections per provider including:
+- Cost per query for baseline and graphrag pipelines
+- Total cost for the specified number of queries
+- Monthly and annual projections at 1K queries/day
+- Provider comparison sorted by total cost
+
+## Example
+```
+cost_estimate 10000 --provider ollama --pipeline both
+```
+
+---
+
+# benchmark
+
+Run the HotpotQA benchmark suite through both pipelines and generate a full evaluation report with F1, Exact Match, Context Hit Rate, and cost analysis.
+
+## Parameters
+- `num_samples` (integer, optional, default=50): Number of HotpotQA questions to evaluate
+- `provider` (string, optional, default="anthropic"): LLM provider
+- `model` (string, optional): Specific model ID
+- `output` (string, optional): File path to save results JSON
+
+## Returns
+JSON with:
+- Per-query results with F1, EM, tokens, cost for both pipelines
+- Aggregate metrics (avg F1, avg EM, context hit rate, win rate)
+- Stratified results by question type (bridge vs comparison)
+- Full text benchmark report
+
+## Example
+```
+benchmark 100 --provider openai --model gpt-4o-mini --output results.json
+```