Reference
doctor
Claude: /rn-dev-agent:doctorCodex: $rn-dev-agent:doctorWhat it reports
Section titled “What it reports”Codex doctor runs the package-local health executable and keeps independent axes for:
- Codex host support (
0.145.0live-refresh floor; older hosts restart-only); - exact plugin ID/version/enabled state;
- selected package materialization and all 27 skills/helpers;
- effective
cdpregistration; - a side-effect-free MCP initialize/
tools/listcontract probe; - direct and caller-observed
proof_captureschema state; - complete/partial caller-observed task skill and MCP inventories;
- prior observed transport/app status;
- linked-worktree action inheritance and repository-local hook status, without exposing private source paths or reading action bodies;
- all findings, a primary finding, and ordered recovery commands.
Unknown or partial task evidence is never classified stale. A direct contract probe is not represented as the active task transport or app state.
When the React Native app root is known, both hosts also run the packaged
read-only session-doctor report probe and classify session-authority health as
green, yellow (a proven-dead owner of this exact root, which the next transport
start releases on its own), or red. A red report names the exact wedge — an
owner whose process identity cannot be proven, a retained startupCleanupBlocked
obligation, or an owner belonging to a different app root — instead of a generic
stale-lock story, and names the supported repair command rooted where it can
succeed. Only a proven-dead owner is ever released; see
recovering a wedged source root.
A task that received zero rn-dev-agent skills cannot invoke doctor. Use the
external exact-version cache bootstrap from Troubleshooting.
/mcp verbose is inventory inspection only.
Main diagnoses
Section titled “Main diagnoses”Missing/disabled installation, corrupt/version-mismatched materialization, missing/disabled MCP registration, contract-probe failure, explicit stale skill or MCP discovery, observed transport closure, server/host schema exposure, observed app disconnection, legacy restart requirement, healthy disk with unknown task state, and healthy.