[AutomatosAI]

What and why

PRD-262 §8 row 16, "fixable Critical / unallowlisted KEV in images we build: 0 / 0", fails on main. images.yml run 39399292631 shows 2 blocking findings in each of two images. The policy runs in report mode, as Gerard decided on 20 Oct (US-004), so nothing failed a build. This PR clears all four.

1. agent-opt-worker: litellm 1.80.0 → 1.91.5 (485ded7)

• The advisories: GHSA-7488-6r32-c95q (High, KEV, fixed in 1.84.0) and GHSA-v4p8-mg3p-g94g (High, KEV, fixed in 1.83.7). Both are in the proxy's MCP endpoints.
• Why 1.80.0 was held: ai-evaluation 1.1.0, the newest on PyPI, caps openai<2, and every litellm from 1.84 on needs openai>=2.20.0,<3. overrides.txt now lifts the cap. The SDK's upstream repo main already declares openai>=1.109.1,<3.
• The version: 1.91.5 is the newest pure-Python wheel and adds no package. 1.92 adds a Rust extension; 1.96 and 1.98 add pydantic-settings, boto3 and botocore.
• The lock: only three pins move: litellm 1.80.0 → 1.91.5, openai 1.109.1 → 2.54.0, importlib-metadata 9.0.1 → 8.9.0.
• Allowlist: the three litellm 1.80.0 Critical entries leave .grype.yaml, since nothing matches them now.
• Compatibility: openai 2.0's only breaking change is the Responses API tool-call output type, which neither SDK uses. litellm 1.84–1.95's breaking changes are all in the proxy, which the worker doesn't use. In a locally built image, the service's suite passes (44) along with the import smoke, and a probe of LiteLLMGenerator.generate, litellm.completion and all four optimizers passes. pip check shows the one expected mismatch: ai-evaluation's openai<2.
• Size: about +11 MB compressed (local estimate). The image is now over its 150 MB budget, which main already exceeds by 0.2 MB. The budget only warns and is not raised here; it's worth revisiting once CI reports the real size.

2. media-render: esbuild 0.25.12 built from source on Go 1.26.10 (2094326)

• The advisories: GO-2026-4337 (crypto/tls, Critical) and GO-2026-6612 (net/http, Critical) are in npm's esbuild 0.25.12 binaries, which are built with go1.23.12.
• Why no bump fixes it: Hyperframes requires esbuild ^0.25.12, including its newest release 0.8.181. The newest esbuild (0.28.2) and esbuild main both build with Go 1.26.5. GO-2026-6612 is fixed in Go 1.26.9 / 1.27.2. esbuild can't be pruned, because the render path calls transformSync.
• The fix: a new esbuild stage on golang:1.26.10-trixie@sha256:9ba7dc6d… runs go install …/cmd/esbuild@v0.25.12 with esbuild's release flags. The js stage copies the binary over npm's platform binary and hard-links esbuild/bin/esbuild to it, and the build fails unless require("esbuild").transformSync runs. The Go toolchain never reaches the runtime. The node-stripped stage and js/prune.txt are unchanged.
• Tests:
  • test_prd262_media_render_esbuild.py: Go ≥ 1.26.9, the built version matching package-lock, the flags and the order of the swap.
  • The in-image tests/test_image_esbuild.py: every esbuild binary's Go build info is ≥ 1.26.9, the two paths are one inode, and the JS API runs. It fails on main's image.
• Local amd64 build: the in-image suite passes 355, including real renders and snapshots. The module trace gives "PASS 12 pruned paths, none loaded". On disk it's 1,296.96 MB against the 1,300 MB bar: +1.25 MB, the Go 1.26 binary.

Local scan: Grype v0.118.0 (the 2026-10-20 database) plus image_scan_policy.py check --mode fail gives agent-opt-worker 0 blocking and media-render 0 blocking.

Test plan

• Static: 372 passed across 22 PRD-262 and media-render files. Ruff and the shape check are clean.
• Local images: agent-opt-worker's suite passes (44) and media-render's in-image suite passes (355). The scan shows 0 blocking for both.
• images.yml on this PR: "agent-opt-worker (amd64): … 0 blocking" and "media-render (amd64/arm64): … 0 blocking". arm64 media takes the @esbuild/linux-arm64 path and wasn't built locally.
• test.yml's agent-opt-worker lane: the lock check, the build, the import smoke, the in-image pytest and the 401 checks.
• media-render.yml dispatched on this branch: the build, the in-image suite with test_image_esbuild, the module trace, the renders, the ≤ 1,300 MB bar and the template shards.

[claude]

⚠️ Code review skipped — your organization has no extra usage available to pay for this review.

If your organization's extra usage balance is empty, an organization admin can add extra usage credits at claude.ai/admin-settings/usage. If its monthly spend limit was reached, an admin can raise it on the same page. If neither applies, contact Anthropic support.

Once extra usage is available, someone with write access to this repository can comment @claude review on this pull request to trigger a review.

[coderabbitai]

Warning

Review limit reached

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Next included review available in 53 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: e5a89b3d-03c8-47a5-9f93-4c34dc28e1d4

📥 Commits

Reviewing files that changed from the base of the PR and between bcc3f79 and 2094326.

⛔ Files ignored due to path filters: 1

• services/agent-opt-worker/requirements.lock is excluded by !**/*.lock

📒 Files selected for processing: 8

• .grype.yaml
• orchestrator/tests/test_prd262_agent_opt_worker_image.py
• orchestrator/tests/test_prd262_media_render_esbuild.py
• orchestrator/tests/test_prd262_media_render_stages.py
• services/agent-opt-worker/overrides.txt
• services/agent-opt-worker/requirements.in
• services/media-render/Dockerfile
• services/media-render/tests/test_image_esbuild.py

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.}

[AutomatosAI]

CI evidence for this PR's head:

• images.yml: agent-opt-worker (amd64): 176 matches, 0 blocking, media-render (amd64): 725 matches, 0 blocking, media-render (arm64): 725 matches, 0 blocking. PRD-262 §8 row 16 is now 0 / 0 for every image we build.
• media-render.yml, dispatched on this branch (run 39408716156): all 5 jobs green (the fixture render and 4 template shards). The in-image suite passes 355, including all 3 test_image_esbuild.py tests (every esbuild binary is built with a patched Go; one binary where npm put it; the JS API runs at the locked version). On disk it's 1296 MB, within its 1300 MB bar: PASS.
• test, images and CodeQL: green.
• smoke-images (non-required): red on the known core-stack step (PermissionError … /workspaces/<c1>/repos), the same as on perf(prd-262): API — W2–W5, pull-first self-host, boot, and the 19 Oct cuts (1,920 → 789 MB) #1226 and perf(prd-262): worker cuts — 1,074 → 910 MB, toolchain kept #1220. That's P256-FIX-WRK-1, not this change.