claude-api

Reference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file).

Agent Skill from anthropics/skills

Copilot Studio

Needs changes

Fails: Description is at most 1,024 characters (CS-DESC-LENGTH)
description is 1,068 characters
rule · Copilot Studio: Troubleshoot skills that don't load (limit value from the Agent Skills specification)

Warnings (don't change the verdict):

Warning: Description is a single line (CS-DESC-ONELINE)
description spans more than one line
rule · Copilot Studio: Troubleshoot skills that don't load

Microsoft 365 Copilot Preview, Frontier tenants only

Needs changes

Fails: Instructions are under 20,000 characters (M365-INSTRUCTIONS)
instructions are 103,964 characters
rule · Microsoft 365 Copilot: What is a skill

Warnings (don't change the verdict):

Warning: No sign of network access, package installs or authenticated calls (M365-SANDBOX)
SKILL.md mentions Python HTTP client, curl/wget
rule · Microsoft 365 Copilot: Script execution sandbox
Warning: No sign of network access, package installs or authenticated calls (M365-SANDBOX)
scripts: shared/preserved-thinking-migration/drop_block_probe.py (Python HTTP client, credentials / authenticated call)
rule · Microsoft 365 Copilot: Script execution sandbox

An agent can have at most 8 skills. Custom skills aren't available in tenants that use Microsoft Purview Information Barriers.

Compatible isn't the same as safe. These verdicts only say whether a skill's files follow Microsoft's documented load rules. The checker reads files and never runs them, and it doesn't review what a skill or its scripts actually do. Read a skill's instructions and scripts before you add it to an agent.

Standards signals

Doesn't meet every agentskills.io required field.

Frontmatter	Present and parsed Spec: frontmatter
name	Meets the spec (1-64 chars, a-z 0-9 hyphens) Spec: name
description	Missing or over 1,024 characters Spec: description
compatibility	Absent or under 500 characters Spec: compatibility
Name matches folder	Yes (`claude-api`) Spec: name

Observed from the public SKILL.md at the commit above on 2026-10-23. Not a certification: AAIF and the Agent Skills project don't certify skills. Validate your own skill with skills-ref validate.

Source repo	anthropics/skills
Path	skills/claude-api/SKILL.md
Commit checked	`dbd4588f9e10`
Repo license	No repo-wide license file · we list metadata only
Skill license file	Apache-2.0
Files	96
Scripts	4 (4 scanned for the sandbox heuristic)
Instructions	103,964 characters
Deepest folder	3 below the skill folder
Installs (skills.sh)	73,645
Rules version	2026-10-22

Re-check the latest version How skill checks work

Is this wrong or out of date? Email us. Skills data built 2026-10-23; rules as of 2026-10-22.

Rule-based checks of public files against Microsoft's documented load rules, not tested in Copilot Studio or Microsoft 365 Copilot. Independent project, not affiliated with Microsoft.