Skip to content

official-docs.md

Latest commit

 

History

History
54 lines (47 loc) · 3.59 KB

official-docs.md

File metadata and controls

54 lines (47 loc) · 3.59 KB
Raw

Official docs

Find a current Claude Code docs page in https://code.claude.com/docs/llms.txt, the platform's self-updating page list, and read it through /discovery:read-docs. Model-behavior pages (prompting, thinking, evaluation) are on platform.claude.com, listed in https://platform.claude.com/docs/llms.txt. Never act on remembered page content: re-fetch the page. The upstream-drift convention owns how a fetched page backs a record, and its fetch route owns how to read one: a summarizing fetch of a long page truncates it and then reports what it never reached as absent, so read the .md channel verbatim and record no verdict from a truncated read.

Rules for citing these pages

  • Prompting doctrine. The instruction text a plugin ships (skill bodies, agent definitions, hook-injected prose, embedded sample prompts) is authored against the platform prompting pages, not the harness pages.
  • Two best-practices pages. The platform page (https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices) is the cross-product Agent Skills guide for skill bodies; the Claude Code page (https://code.claude.com/docs/en/best-practices) is the harness guide for CLAUDE.md, permissions, and sessions. They share only their slug, so cite the one you mean by its full URL.
  • The changelog. The prose pages can lag the changelog: a behavior can change in a release and reach the topic page a release or more later, and when the two disagree the changelog is the one that matches the running harness. The rendered page summarizes; fetch the raw markdown (…/changelog.md, or the upstream CHANGELOG.md) when you need a byte-exact version-pinned quote, because a summarizing fetch of this page has produced inconsistent readings of the same entries. A changelog citation records what changed in a version, so always pin the version, and pair it with the topic page rather than replacing it, since the topic page stays authoritative for mechanism and semantics.
  • JSON Schemas. CI validates against these by URL and never treats them as a load-time contract: marketplace.json → https://json.schemastore.org/claude-code-marketplace.json, plugin.json → https://json.schemastore.org/claude-code-plugin-manifest.json (published on SchemaStore, sourced from the same plugin system the docs describe). plugin.json files carry no $schema key: claude.ai's marketplace sync strips it with a warning, and Claude Code ignores it at load time.

Evaluation guidance (platform docs)

This marketplace authors model-graded eval fixtures for its skills (see the migration playbook's eval warrant policy) and ships the evals plugin distilling this guidance.

Page Official doc page As of
Define success criteria and build evaluations https://platform.claude.com/docs/en/test-and-evaluate/develop-tests 2026-08-08
Evals cookbook (source: anthropics/claude-cookbooks misc/building_evals.ipynb) https://platform.claude.com/cookbook/misc-building-evals 2026-08-08

Each row's As of date is when the page was last read live; a fetch that no longer matches a row is that row's recheck trigger.