mirror of
https://github.com/borghei/Claude-Skills.git
synced 2026-09-14 19:41:02 +08:00
5e4f0f8a9c
Audit of the library's AI/LLM skills found near-zero coverage of the newest agentic features. Adds 4 skills and upgrades 4, taking the library to 343 skills / 17 domains / 786 Python tools (engineering 82 -> 86). New skills (each: stdlib-only tools + reference knowledge bases): - extended-thinking-architect — reasoning-effort / thinking-budget decisions - batch-api-orchestrator — batch vs realtime, bulk job design - computer-use-automation — GUI/browser agent loop, reliability + safety - agentic-evaluation-framework — LLM-as-judge rubrics + pairwise ranking Enhanced: - llm-cost-optimizer — prompt caching / batch / reasoning-effort economics - context-engine — memory tool, context editing/compaction, 1M-context - agent-workflow-designer — subagent scoping + multi-model cost estimation - mcp-server-builder — streaming, batch, computer-use tool patterns All content is model-agnostic and principle-focused (no hardcoded prices or invented API signatures); all tools are stdlib-only with --json output and were compiled + run. Manifests, website (343 skill pages), and prose regenerated.