Feature: consult command for persistent multi-turn Q&A sessions

[luqmanbello]

Problem

The current commands are all single-shot: review, adversarial-review, rescue. There's no way to have an ongoing conversation with Codex about a topic -- exploring a design, asking follow-up questions, iterating on an approach -- while maintaining context across turns.

The plugin already has thread management infrastructure (thread/start, thread/resume). Exposing this as a user-facing command would unlock conversational use cases.

Proposal

Add /codex:consult [topic] command:

• Starts a new Codex thread scoped to the current repo
• Subsequent /codex:consult [follow-up] resumes the same thread
• --fresh flag starts a new thread (discards previous context)
• Session mapping persisted per-repo so threads survive across commands
• No structured verdict required -- response presented verbatim

Use cases:

• "How does the auth middleware work in this repo?" -> follow-up: "What would break if we replaced it with JWT?"
• Iterating on a design: ask, get feedback, refine, ask again
• Exploring unfamiliar code with a knowledgeable companion

Implementation notes

The broker already supports thread/start and thread/resume. This would be a new command definition + prompt template + session state tracking (repo path -> thread ID mapping).

[dkundel-openai]

How would this be different than using /codex:rescue? It should by default suggest to use the same thread or you can use /codex:rescue --resume [follow-up] to be more explicit

[luqmanbello]

Great point -- from my own implementation as a skill they do overlap mechanically since both use threads under the hood. The difference is really about what I am trying to do with it.

With rescue, I'm handing off work -- "fix this test", "refactor this function". Codex has write access and the thread lives as long as the task needs it. Once it's done, I move on.

What I'm describing with consult is more like thinking out loud with Codex. I want to ask "how does the auth middleware work here?", get an answer, then follow up with "ok what would break if we replaced it with JWT?" -- and have that context carry across turns. No file modifications, just back-and-forth exploration.

From exploring the plugin I notice I could do this with rescue --resume, but it feels off a bit It's like I am giving Codex write access when all I want is a conversation. A dedicated command would make the intent clearer and keep things read-only by default.

That said, if the team feels rescue --resume covers this well enough, maybe it's just a matter of documenting that pattern more visibly?