fix(http): disable HTTP/2 to fix undici decompression issues

[shamashel]

Fixes openclaw/openclaw#3637

Summary

When undici fetches compressed content over HTTP/2, it fails to decompress the response, causing JSON parse errors:

✖ Unexpected token '',"..."... is not valid JSON

This change disables HTTP/2 (allowH2: false) to force HTTP/1.1 which handles compression correctly.

Changes

• packages/clawdhub/src/http.ts: Changed allowH2: true → allowH2: false

Testing

• Verify clawhub CLI works with compressed responses
• Check that skill search/install operations succeed

Context

The undici HTTP/2 implementation does not properly decompress response bodies when servers return compressed content over HTTP/2. This causes the CLI to receive binary data instead of valid JSON.

See: openclaw/openclaw#3637

[vercel]

Someone is attempting to deploy a commit to the Amantus Machina Team on Vercel.

A member of the Team first needs to authorize it.

[vercel]

Additional Suggestion:

e2e test file uses allowH2: true while production code uses allowH2: false, causing e2e tests to experience undici decompression issues

[steipete]

Closing as obsolete.\n\n- already removed (see entry in released 2026-02-13: "HTTP: remove from Undici agent to prevent on Node.js 22+").\n- Undici v7 default is anyway, so the effective fix is already in place by not setting it.\n- PR is currently merge-conflicting; rebasing just to land a redundant change doesn’t seem worth it.\n\nIf the original issue still reproduces on CLI >= 0.6.1 (2026-02-13), let’s capture a minimal repro + add a regression test focused on compressed JSON responses (HTTP/1.1 + HTTP/2) instead of toggling H2.

[steipete]

Follow-up (previous comment lost some inline code formatting due to shell backticks):

Closing as obsolete.

• main already removed the Undici allowH2 override (CHANGELOG.md -> 0.6.1 - 2026-02-13 -> Fixed -> "HTTP: remove allowH2 from Undici agent to prevent fetch failed on Node.js 22+").
• Undici v7 default is allowH2=false, so the effective fix is already in place by not setting it.
• PR was merge-conflicting; rebasing to land a redundant change doesn’t seem worth it.

If the original issue still reproduces on CLI >= 0.6.1 (2026-02-13), suggest: minimal repro + regression test around compressed JSON responses (HTTP/1.1 + HTTP/2).