# Primary Sources

- [Anthropic Messages Token Count API](https://platform.claude.com/docs/en/api/messages-count-tokens) - defines `POST /v1/messages/count_tokens`, accepted request fields, and the `input_tokens` result.
- [Anthropic token counting guide](https://platform.claude.com/docs/en/build-with-claude/token-counting) - states that token counting supports system prompts, tools, and extended thinking; counts may include Anthropic-added system optimization tokens; the result is an estimate; and the endpoint is free subject to a separate requests-per-minute limit.
- [Anthropic Claude Agent SDK overview](https://platform.claude.com/docs/en/agent-sdk/overview) - identifies the Agent SDK as the programmable Claude Code agent loop.
- [Anthropic Agent SDK TypeScript reference](https://platform.claude.com/docs/en/agent-sdk/typescript) - defines the optional `tools`, `systemPrompt`, and `settingSources` query options used to describe the captured configuration.
- [Anthropic Agent Skills in the SDK](https://platform.claude.com/docs/en/agent-sdk/skills) - documents progressive disclosure: skill metadata is discovered at startup and full skill content loads when triggered.
- [`@anthropic-ai/claude-agent-sdk@0.3.168`](https://www.npmjs.com/package/@anthropic-ai/claude-agent-sdk/v/0.3.168) - exact SDK package measured by the capture script.
- [Piebald-AI/claude-code-system-prompts](https://github.com/Piebald-AI/claude-code-system-prompts) - comparison catalog; the exact inspected commit is recorded in `capture-provenance.json`.