docs(factories): document agentDefaults.computerUseModel (unreleased feature) - #581
docs(factories): document agentDefaults.computerUseModel (unreleased feature)#581warp-agent-staging[bot] wants to merge 3 commits into
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
There was a problem hiding this comment.
Factory review
The computerUseModel documentation is accurate and complete: it covers the Warp Agent harness gate, behavior when computer use is disabled, compatibility with harness, restricted model guidance without publishing an allowlist, worker-version guidance, and the agent and automation overrides.
Blocking finding
The PR has no visual proof of the rendered user-facing documentation change. Capture a preview screenshot or video demonstrating the new section, then attach it to the PR description.
Validation
- GitHub CI: build, link, audit, analysis, CodeQL, Vercel Preview, and CLA checks passed.
npm run typecheckpassed with 0 errors.- Internal link checker passed with 0 broken links.
style_lint --changedreported only one unrelated pre-existing glossary warning../script/presubmitis not present in this repository. The local build was not accepted as a pass because the sandbox provides unsupported Node 25 and failed while prerendering the unrelated/_llms-txt/support.txtroute; the authoritative GitHub build check passed.
Verdict: Changes requested
Task record: ticket_id=364c959b-3c3d-45ee-88df-9c3efe16301a, ticket_source=adhoc, task_id=364c959b-3c3d-45ee-88df-9c3efe16301a, factory_name=Doc 'n Roll', requester daniel@warp.dev.
Review run: https://oz.staging.warp.dev/runs/01a01bbb-79f3-7d5e-895d-fcfcb8eeaff8
There was a problem hiding this comment.
Factory re-review
The sole prior finding is closed. I directly inspected the three embedded screenshots and confirmed that they render:
- The complete
agentDefaults.computerUseModelsection, including its YAML example, all behavioral caveats, and both links. - The
agents/<name>/agent.mdfrontmatter override bullet containingcomputerUseModel. - The automation Execution overrides text containing
computerUseModel.
The screenshots show no raw markup, error state, broken layout, or missing required content. The prior review thread is resolved, the documentation diff has not changed, and all GitHub checks remain successful.
Verdict: Accepted
Task record: ticket_id=364c959b-3c3d-45ee-88df-9c3efe16301a, ticket_source=adhoc, task_id=364c959b-3c3d-45ee-88df-9c3efe16301a, factory_name=Doc 'n Roll', requester daniel@warp.dev.
Review run: https://oz.staging.warp.dev/runs/01a01bbb-79f3-7d5e-895d-fcfcb8eeaff8
|
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 Powered by Oz |
There was a problem hiding this comment.
Overview
This PR documents the new agentDefaults.computerUseModel key on the factory-as-code reference page and adds it to the agent and automation override lists. The change is limited to docs copy and key lists; I found one terminology issue in the new section.
Concerns
- The new
agentDefaults.computerUseModelsection usessubagentfor the computer use runtime, but the docs glossary reservessubagentfor child agents created by a parent agent. The inline suggestion keeps the description focused on the user-visible configuration behavior. - The PR description says this feature is unreleased, so keep this draft unmerged until the product surface ships. This is a release-readiness constraint, not a counted diff finding.
Verdict
Found: 0 critical, 0 important, 1 suggestions
Approve with nits
Comment /warp-agent-review on this pull request to retrigger a review (up to 3 times on the same pull request).
Powered by Oz
There was a problem hiding this comment.
Factory re-review
The documentation change builds and renders correctly, and the existing visual proof covers the new section plus both override lists. One unresolved terminology finding still needs a copy-only correction before acceptance.
Finding
src/content/docs/factories/factory-as-code.mdx:213— Replace “computer use subagent” with wording that describes computer use without redefiningsubagent, which the glossary reserves for child agents created by a parent agent.
Validation
- GitHub CI checks: passed.
npm run typecheck: passed with 0 errors.npm run build: passed.- Internal link checker: passed with 0 broken links.
- Live local render: verified the new section, both override lists, and both added links with no console or layout errors.
- Local
npm run lint: unavailable becausetrunkis not installed in this environment;./script/presubmitis not present in this repository.
Verdict
Rework needed
Review run: https://oz.staging.warp.dev/runs/01a04013-cc87-763e-88ea-d23f55833786
Computer use runs as part of the agent's own run, not as a separate child agent, so 'subagent' conflicted with the terminology glossary's definition of a subagent as a child agent created by a parent agent. Co-Authored-By: Warp <agent@warp.dev>
There was a problem hiding this comment.
Overview
The PR documents agentDefaults.computerUseModel and lists the override on agent and automation definitions. The final rework replaces both inaccurate uses of “computer use subagent” with “computer use,” matching the glossary.
Verdict
Checks: CI ✅ · build ✅ · typecheck ✅ · internal links ✅ · visual proof ✅ · style lint: no new diff issues (npm run lint unavailable because trunk is absent; no ./script/presubmit in this repository)
Found: 0 critical, 0 important, 0 suggestions
Accepted
Review run
https://oz.staging.warp.dev/runs/01a0402f-2f45-7e82-b5e3-1ddde6fc59fe
Summary
Documents the new
computerUseModelfactory-definition key on the "Factory definition syntax" page (docs.warp.dev/factories/factory-as-code/), per the drafted copy from Doc 'n Roll intake.computerUseModelis implemented inwarpdotdev/warp-server#15542andwarpdotdev/warp#15289, plus a pendingoz-agent-workerfollow-up. This PR is intentionally opened as a draft and must stay a draft until all three have shipped, so the docs don't publish ahead of the feature. A human should promote it to ready for review only after that.Changes
In
src/content/docs/factories/factory-as-code.mdx:### agentDefaults.computerUseModelsubsection afteragentDefaults.workerHost, describing the key, its example usage, its scope (Warp Agent harness only, ignored when computer use is disabled), its non-exclusivity withharness, and the self-hosted worker version requirement. Includes a one-line pointer to model choice for agents noting computer use supports a restricted model set (no specific model IDs enumerated, since the allowlist changes over time).computerUseModelto theagents/<name>/agent.mdfrontmatter key list (the "same keys asagentDefaults" bullet), so agents can override the factory default.computerUseModelto theautomations/<name>/automation.mdexecution-override key list, with the same semantics.Edit 4 from the request (pointer from the new section to a "model choice for agents" page) is folded into edit 1 above, since that page already exists at
/agents/inference/model-choice/.Verification
This is a documentation-copy-only change (new prose, a YAML example, and additions to existing key-enumeration lists) with no runnable behavior — pure data/copy, one of the testing-exempt categories in
factory-verification, so no regression test was added.Ran the repo's documented checks:
npm install && npm run build— succeeds, no errors (Complete!, 376 docs generated, sitemap + search index built).python3 .agents/skills/style_lint/style_lint.py --changed— 0 errors. One pre-existing warning (UNRECOGNIZED-TERMfor "GitHub" bolded on line 20) is unrelated to this diff and out of scope for this change.Visual verification
This is a rendered docs page, a user-facing surface, so it needs visual proof per
factory-verification/factory-ui-verification. The proof is embedded below, captured from a livenpm run devrender of this branch (via a computer-use session) — the three sections shown are the newagentDefaults.computerUseModelsection, theagents/<name>/agent.mdfrontmatter key list, and theautomations/<name>/automation.md"Execution overrides" list, each showingcomputerUseModelrendering correctly.Computer-use screenshots (3)
Area 1: The "agentDefaults.computerUseModel" section fully visible, showing the heading, all three paragraphs, the syntax-highlighted YAML code block, and inline links ("model choice for agents", "self-hosted worker").
Area 2: The frontmatter key bullet list under "agents/<name>/agent.md" heading, showing all four bullets including the one enumerating model, harness, runner, environmentId, secrets, mcpServers, workerHost, computerUseModel.
Area 3: The "Execution overrides" section heading and paragraph enumerating the keys: model, harness, runner, environmentId, secrets, mcpServers, workerHost, and computerUseModel.
Rework changes (rework #2)
agentDefaults.computerUseModelsection (the model-summary sentence and the harness/exclusivity paragraph), matching the terminology glossary's definition ofsubagentas a child agent created by a parent agent. Applied the exact suggested wording from the review comment.Originating thread: adhoc Factory MCP intake (task
364c959b-3c3d-45ee-88df-9c3efe16301a, factory "Doc 'n Roll", requester daniel@warp.dev)