#12 · stacks/uplink

Part of #4. Last in the order, because it documents what the others built.

.env.example was updated when the engine keys landed in #7. The README was not.

README.md#L3 and #L6 say Uplink "runs each one through Claude Code".
The architecture diagram at README.md#L22 shows claude -p ... --resume <thread session> as the only path.
README.md#L43 and #L66 make claude setup-token and CLAUDE_CODE_OAUTH_TOKEN the only sign-in.
README.md#L112 describes "the Claude Code engine (engine.ts)", "Conversation (a thread and its Claude session)" and "a stub claude".
The command table at README.md#L104 does not mention the new control words.

Reword the opening so the product is "your agent" rather than Claude specifically, while keeping Claude as the default, because that is what it is.
Redraw the diagram to show the engine as a choice at the point where the CLI is spawned.
Document both sign-in paths, and that Codex needs no API key: codex login --device-auth or codex login --with-access-token.
Add claude and codex to the control words table and to the layout section's description of commands.ts, once those land.

Do not oversell it. If Codex support ships with a confirmed-but-narrow set of behaviours, say what works rather than implying parity.
The Layout section is a map people use to find code. Keep it accurate to the files that exist after the other issues land.
House rule: no em-dashes or en-dashes.

A reader who has only Codex can install, sign in and text it using the README alone.
The Layout section names files that exist.