Where the agents guide is prose, the command index is structured: two JSON catalogs an agent can load and enumerate directly, so it never has to guess a verb or the meaning of an error. Both come straight from the installed binary, and both are the same source the generated command reference is built from.

The command catalog

strata agents commands --json returns the command catalog — each command with its access class, batching and commit behavior, description, docs path, and the error codes it can raise:

strata --cache agents commands --json

{"data":{"command_count":32,"commands":[{"access":"write","batch":"itemwise", ... }]}}

Each entry is fully described: its path, access mode (read/write), batch semantics, commit behavior, response model, and the exact error codes it can return. An agent can route on those fields — for example, only calling write commands inside a branch it forked, or knowing in advance which codes a command may raise.

The error registry

strata agents errors --json returns the public error registry — every code with its class, hint, retry policy, and reference URL:

strata --cache agents errors --json

{"data":{"count":204,"errors":[{"class":"invalid_argument","code":"invalid_argument.engine.branch_catalog","commit_outcome":"not_started","hint":"Correct the invalid field named by the error message and retry the operation.","message":"The request contains invalid input.","ref":"https://stratadb.org/e/invalid_argument.engine.branch_catalog","retry_policy":"never"}, ... ]}}

This is the same registry that backs the browsable /e/<code> pages and the error reference. Recover by code, never by message text — the message is for humans and may change; the code is stable.

One source, many renderings

The command index is the IDL: the machine-readable contract for every command. The generated command reference on this site, the CLI’s own help, and the MCP tool schemas are all renderings of it. Because they share one source, they cannot drift — if a page and the binary ever disagree, the binary wins, and you should check that your installed version matches these docs.

Related

• For AI agents — the section overview.
• The agents guide — the prose counterpart.
• Machine-readable docs — the website’s own machine surface (llms.txt, .md mirrors, the /e/ registry).
• Command reference — the human-readable rendering, per family.