feat(fetch-use): default experimental.fetch_use to off

[Alezander9]

Summary

Eval runs showed routing webfetch through fetch-use had no measurable impact on agent accuracy versus the native HttpClient path, while consistently adding latency on every request. Flip the default so the proxy is opt-in.

Changes

• packages/opencode/src/tool/webfetch.ts: useFu now requires experimental.fetch_use === true (was !== false).
• packages/opencode/src/config/config.ts: schema description updated to reflect default-false and the latency tradeoff.
• packages/bcode-browser/skills/BROWSER.md: skill instructs the agent how to opt in rather than how to opt out.
• packages/bcode-browser/test/fetch-use.test.ts: header comment updated (opt-in, not opt-out). No test behavior change.

The feature itself is unchanged — only the default. Users who want the Chrome TLS fingerprint + residential proxy set experimental.fetch_use: true in opencode.json.

Testing

• bun typecheck clean in packages/opencode and packages/bcode-browser.
• No existing tests assert the default-on behavior (the fetch-use suite covers the layer/live path; the config-gated branch in webfetch.ts wasn't covered by a unit test in either direction).

[cubic-dev-ai]

No issues found across 4 files

[cubic-dev-ai]

1 issue found across 1 file (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/bcode-browser/skills/BROWSER.md">

<violation number="1">
P2: This says fetch-use is automatic when `BROWSER_USE_API_KEY` is set, but it still requires `experimental.fetch_use: true` to take the proxy path.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review, or fix all with cubic.}