token-engineering

Plugin token-engineering from marketplace rahul-resources in Rahul-s-007/resources. Commands read from the repo's manifest — 1 month ago.

Skills (6)

• batch-planner

  Decide whether an LLM workload fits Batch API or Flex tier pricing (about 50% off) and design it correctly for Anthropic, OpenAI or Gemini. Covers custom_id joins, out-of-order results, partial expiry and resubmission, async validation, batch plus caching, and flex timeouts and fallbacks. Use when s

• cache-audit

  Audit code that builds LLM prompts for silent prompt-cache busters such as timestamps or IDs in the system prompt, unsorted JSON, unstable tool order, per-user content before shared content, history truncation, non-deterministic RAG order, and cache-key parameter changes. Use when cache hit rate is

• cache-setup

  Implement or fix prompt caching for a specific provider. Covers Anthropic cache_control breakpoints and TTL, OpenAI implicit caching, prompt_cache_key and prompt_cache_retention, and Gemini implicit vs explicit cachedContents. Use when adding caching to an LLM integration, choosing breakpoint placem

• cost-estimate

  Estimate and compare LLM API costs with caching, batch and long-context pricing effects across Anthropic, OpenAI and Gemini, using prices fetched live from the official pricing pages. Use for monthly cost projections, break-even on cache writes vs reads (5 min vs 1 h TTL), checking whether a route c

• prompt-fingerprinting

  Measure prompt stability without logging prompts. Hash each prompt segment and a running prefix hash, find which segment breaks the cache between requests, choose breakpoint positions from stability profiles, and replay logged traffic to predict cache hit rate for 5 min, 30 min and 1 h TTLs. Use whe

• token-engineering

  Entry point for reducing LLM API spend. Use when the user mentions LLM costs, token bills, prompt caching, cache hit rate, batch API, flex tier, long-context pricing, or asks to "optimize" or "audit" prompts for Anthropic, OpenAI or Gemini. Diagnoses what they need and routes to the right sub-skill.

Similar plugins