Build, debug, and migrate applications using the Claude API and Anthropic SDKs.
npx skills add https://github.com/irrit-us/agent_misc --skill claude-api-irrit-us
Please help me install this Agent Skill. Skill: claude-api Source: https://github.com/irrit-us/agent_misc/tree/main/skills/claude-api Command: npx skills add https://github.com/irrit-us/agent_misc --skill claude-api-irrit-us
High-intent search queries and answers about installing and using this skill.
Set cache_control with type ephemeral on the last block of your system prompt or use top-level auto-caching on messages.create. Keep stable content first in the prefix and verify hits via usage.cache_read_input_tokens; zero reads across repeated requests indicates a silent invalidator.
Use the exact model ID strings from the current models table, such as claude-opus-4-8, without date suffixes. Replace deprecated budget_tokens with adaptive thinking and remove assistant prefills, which return 400 errors on Opus 4.6 and later.
Python, TypeScript, Java, Go, Ruby, and PHP support beta tool runners with automatic loop handling. C# supports tool use through the Messages API with raw JSON schema definitions but has no class-annotation-based tool runner.
No, Managed Agents is first-party only and unavailable on Bedrock, Vertex AI, or Microsoft Foundry. For agents on third-party providers, use the Claude API with tool use and run the agent loop yourself.
Opus 4.7 and 4.8 removed budget_tokens entirely along with temperature, top_p, and top_k; only adaptive thinking is accepted. On Opus 4.6 and Sonnet 4.6 budget_tokens is deprecated but still functional as a transitional escape hatch.
liuyu520Strip stale thinking.signature blocks from API request copies to prevent 400 errors.
ndestatesMonitors xAI Grok session token usage and prompt-cache hit rates with cost estimates.
GGEduRoute LLM tasks to cost-appropriate models and track cumulative API spend.
davila7Build LLM applications with agents, chains, RAG pipelines, and vector store retrieval.
