Resources · codebook
A tabbed reference to how Claude actually works: every surface, skill, connector, memory scope, and scheduled task, and what each is for.
Claude is one assistant reachable through several surfaces, powered by a family of models, extended by skills, connectors, memory, and scheduled tasks. Most confusion comes from not knowing which piece does which job, or where your context is standing by.
This manual lays all of that out in one place. Seven tabs, each a dense set of tables and a diagram. It is fact-checked against support.claude.com and platform.claude.com, and the footer shows when the sources were last checked.
Start on Overview. It has a three-zone diagram (your devices, Claude's cloud, your tools), a one-table map of every piece, and a "Which piece does the job?" table that says what to reach for and why not the others.
SKILL.md anatomy diagram, the get/trigger/update/scale lifecycleClaude changes monthly, so a static write-up goes stale fast. The manual is a published page with a stable link, and a monthly scheduled task re-checks Anthropic's docs and republishes it in place. The Scheduled Tasks tab documents that exact pattern ("Living-document refresh") and points at this manual as the example.
The Connectors tab carries a stale-fact alert for a rule that used to be true and isn't anymore. That is the point: the page is meant to be corrected, not preserved.
Open the Surfaces tab and read the matrix row for Dispatch next to the row for Cowork. One needs your desktop awake; one doesn't. That single distinction is the most common surprise.
Then go to Topic Index and type memory into the filter. Every entry that mentions it stays; everything else drops away. Click one and it jumps you to the right tab, sub-tab, and row.