/consult

Open or continue a Codex consultation thread. Each workspace gets its own thread mapping — calling /codex:consult again from the same repo automatically resumes the prior conversation, so you can have a continuous design discussion across multiple invocations.

Use this when:

• You want to think through a design decision out loud and get a second opinion.
• You want to explore unfamiliar code with a knowledgeable companion ("how does the auth middleware actually work?").
• You're iterating on an approach: ask, get feedback, refine, ask again.

Don't use this when:

• You want code applied — switch to /codex:delegate.
• You want a code review — switch to /codex:review / /codex:adversarial-review.
• The question is one-shot and you don't need a persistent thread — /codex:rescue is fine.

Forwarding rules:

• Use exactly one Bash call to invoke node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.mjs" consult ....
• Default to foreground (this is interactive — turn-by-turn back-and-forth is the point). Honor --background only if the user explicitly asks.
• Strip routing flags (--fresh, --background, --wait, --model) from the question text. They go to codex-companion, not into the question.
• If the user passes --fresh, forward it; the companion will clear the workspace's thread mapping and start a new thread.
• If the user did not pass --fresh and there's an existing thread for this workspace, the companion resumes it automatically.
• Leave --model unset by default. Only forward --model <name> if the user explicitly asks.
• If the user asks for spark, map to --model gpt-5.3-codex-spark.
• Return the Codex output verbatim — don't paraphrase, don't summarize. The user is having a conversation; let Codex's voice come through.
• If the helper reports Codex is missing or unauthenticated, stop and tell the user to run /codex:setup.

Raw user request:

$ARGUMENTS

Examples:

/codex:consult how does the auth middleware actually work in this repo?
/codex:consult what would break if we replaced session tokens with JWT?
/codex:consult --fresh I want to explore caching strategies — different topic
/codex:consult --model gpt-5.5 dig deeper on the JWT migration cost

Refs Linear SUP-373 (W3.2). Implements cc#7 (stale at upstream for 1.5 months).