release: v0.1.1242

[kwakayama]

Ships six fixes that have been stranded on main.

Why they were stuck

0.1.1241 was published at ~11:00 UTC today, and main stayed at 0.1.1241. Every fix merged after that point could not reach any environment — the release job reads deno.json, sees no version change, and correctly does nothing.

src/utils/version.test.ts asserts that VERSION matches deno.json, not that it increased. So main can accumulate merged fixes at an already-published version indefinitely with CI fully green. That is worth fixing separately; this PR just unblocks what is queued.

What ships

PR	Fixes
#3817	Skip Azure content-filter preamble frames — issue-inbox#543
#3818	Route Azure GPT models through chat — issue-inbox#548 (gpt-5.4 / gpt-5.5 are 100% broken today)
#3820	Replay provider metadata across tool steps — issue-inbox#549 (Gemini 3.x tool calls)
#3821	Preserve every signed Gemini tool step
#3826	Withhold replay metadata when a tool call was suppressed
#3822	Record upstream status on failed-closed metadata lookups
#3823	Surface platform errors from durable run setup as their real status
#3799	Publish the embedded root route once, never as a dotfile
#3796	Report ignored feature file trees

#3817 and #3818 must ship together. Routing Azure GPT models to Chat Completions is what makes Azure's prompt_filter_results preamble reachable, and #3817 is what stops that preamble failing the stream parser. Shipping #3818 alone would trade a 404 for invalid successful stream (empty choices event had no usage).

Impact once deployed

This is what unblocks the model catalogue. Measured on staging today, 32 real agent runs (16 models × text + tool): 26 passed, 6 failed, and every failure is fixed by a PR in this release.

Model	Today	Fixed by
gpt-5.4, gpt-5.5	404 on every run, text and tool	#3818
gemini-3.1-pro-preview, gemini-3.5-flash	text passes, every tool call 400s	#3820 / #3821 / #3826

Follow-up required — this release alone is not enough

veryfront-agent needs a separate pin bump. The release dispatches veryfront-code-released to veryfront-server, veryfront-job-runner and veryfront-sandbox — but not veryfront-agent, which pins its own npm copy in package.json. Two runtimes execute agent runs, and today only one of them auto-updates. Earlier today Kimi stayed broken on veryfront-agent for 20 minutes after veryfront-server had the fix, which is how this gap was found.

So: merge → publish → veryfront-server deploys automatically → manually bump veryfront-agent (all five veryfront / @veryfront/ext-* packages move together; frameworkDependency.test.ts enforces it) → re-run the model matrix.

Verification

src/utils/version.test.ts — 1 passed (21 steps). All three version files agree at 0.1.1242: deno.json, src/utils/version-constant.ts, and the prebundled hydration runtime that embeds the version string. 0.1.1242 is above the published 0.1.1241.

[github-actions]

📦 Client bundle boundary

Entrypoint	Modules	Source size	Server leaks
src/index.client.ts	325	1937 KiB	✅ 0

A server module in a client graph aborts hydration in the browser. New leaks fail CI; known leaks are tracked in scripts/lint/client-bundle-baseline.json to burn down.

[coderabbitai]

Important

Review available on request

• 🔍 Trigger review

Reviews should be triggered manually for repositories with fewer than 10 stars. Select Trigger review above or comment @coderabbitai review to review the latest changes. For a full review, comment @coderabbitai full review.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c45736e6-ea18-4c16-9eb3-79ffbb338f03

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}