docs: document harness-specific model IDs for oz agent run-cloud

[hongyi-chen]

Summary

oz agent run-cloud --model accepts harness-specific model IDs when running on the claude or codex harness, but the CLI reference described --model as if it only ever took a Warp model ID. This PR documents that, and fixes a duplicated block in the same flag list.

Found by the missing_docs drift-watch audit (changelog item #14011) and verified against crates/warp_cli/src/model.rs.

Scope note

This PR originally also documented the named-agent --prompt / --remove-prompt flags. #387 already covers that, so those hunks have been removed to avoid competing edits to the same lines. See the note below for one correction worth folding into #387.

Changes

src/content/docs/reference/cli/index.mdx

Harness-specific model IDs

• Added a Selecting a model for a third-party harness subsection with Claude Code and Codex examples.
• Clarified in both --model bullets that oz model list only covers the built-in oz harness, and that an unrecognized ID is rejected by the harness rather than by Warp.

Duplicated flag block

• Removed a duplicated --harness / --claude-auth-secret / --codex-auth-secret block from the run-cloud flag list. The same three flags were listed twice with different wording. Not an audit finding, just adjacent to the lines being edited.

Source verification

ModelArgs in crates/warp_cli/src/model.rs:

For third-party harnesses (--harness claude or --harness codex), this sets the harness-specific model. The value is passed directly to the harness. See third party harness docs for list of accepted values.

ModelCommand::List is likewise documented as listing models "for the Warp Agent harness" only.

Note for #387

While verifying the base-prompt flags I traced where the stored prompt actually goes. It is added to the system prompt as an identity-and-behavior section, and does not replace the per-run prompt:

• resolveAgentIdentityPrompt feeds promptParams["agentIdentityPrompt"] in warp-server's primary agent.
• resolveAgentIdentityPromptForHarness prepends # Identity and Behavior / "The user has provided the following instructions on how you should behave:" for third-party harnesses.

#387's phrasing ("the base prompt applied to the agent's runs") could be read as the run prompt. Worth a sentence there clarifying that each run still supplies its own task.

Verification

• npm run build passes.
• Internal link check passes (0 broken links).

Deferred findings from this audit run

Run-wide deferrals are listed in the companion audit-bookkeeping PR.

Conversation: https://staging.warp.dev/conversation/53c4f3f8-39bb-46c1-a5c0-66b467db0439
Run: https://oz.staging.warp.dev/runs/019fb91e-74a0-73f4-ac72-f2d88c32fab5

This PR was generated with Oz.

Co-Authored-By: Oz oz-agent@warp.dev

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
docs	Ready	Preview	Jul 31, 2026 11:50pm

[oz-for-oss]

@hongyi-chen

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[oz-for-oss]

Overview

This PR updates the CLI reference to clarify that oz agent run-cloud --model accepts harness-specific model IDs for Claude Code and Codex, while oz model list applies to the built-in oz harness. It also removes a duplicated third-party harness flag block from the same option list.

Concerns

• The new terminal command example should use the docs-standard bash language identifier instead of sh.

Verdict

Found: 0 critical, 0 important, 1 suggestions

Approve with nits

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[oz-for-oss]

💡 [SUGGESTION] Use bash for terminal command fences so syntax highlighting matches the docs style guide.

Suggested change

	```sh
	```bash