:bug: failure to docker build the serverless harness

[maia-iyer]

Describe the bug

PR #155 is failing upon CI and also when I docker build i get error:

% docker build --load -t dev.local/serverless-harness:local .
[1/2] STEP 1/22: FROM node:24-alpine AS builder
[1/2] STEP 2/22: RUN apk add --no-cache build-base python3 pkgconfig pixman-dev cairo-dev pango-dev jpeg-dev giflib-dev librsvg-dev
--> Using cache 28b3e5c2481c8f0fdb45df81924cf66a924231423f0c86163621eeaa29837177
--> 28b3e5c2481c
[1/2] STEP 3/22: RUN corepack enable && corepack prepare pnpm@9 --activate
--> Using cache acb8527728a4ca162303b8e5c3b6c1d40310c8e55e9d35354b6cd2a8032c1a67
--> acb8527728a4
[1/2] STEP 4/22: WORKDIR /app
--> Using cache 79ba5abe2391f53f48d047e9bbfd05ba8c262ae5f7d7e21baa79aca047d7e79c
--> 79ba5abe2391
[1/2] STEP 5/22: COPY pnpm-workspace.yaml pnpm-lock.yaml ./
--> Using cache 79d4b3d389a7ca5853f9d825e4fb4e97fafb81860a314a112b4995070097a595
--> 79d4b3d389a7
[1/2] STEP 6/22: COPY packages/session-backend/package.json ./packages/session-backend/
--> Using cache c5d2abfade0291b1610f78bf713682545a4db045b770220d9a93b04f92b9b2d0
--> c5d2abfade02
[1/2] STEP 7/22: COPY packages/k8s-sandbox/package.json ./packages/k8s-sandbox/
--> Using cache 4fe6d76077d1d5eb4d20edc34c16394ae75bdf4eea2170410b0e1165a3f88c14
--> 4fe6d76077d1
[1/2] STEP 8/22: COPY packages/knative-server/package.json ./packages/knative-server/
--> Using cache 13ed6d83051c7c781d24e744fa1584047391f75ed2d7124fa15f5784fb1cf8e2
--> 13ed6d83051c
[1/2] STEP 9/22: COPY packages/sandbox-relay/package.json ./packages/sandbox-relay/
--> Using cache b7cc3c1b38347b7d385edfa4baaac94dffd18e6335bbb65e0edc70db3a7a672b
--> b7cc3c1b3834
[1/2] STEP 10/22: COPY packages/work-queue/package.json ./packages/work-queue/
--> Using cache ce11fa81b4d29ad83836ae2658772faf94a0fb0baf47aad56c01ccdc326137e7
--> ce11fa81b4d2
[1/2] STEP 11/22: COPY packages/ibac-stub/package.json ./packages/ibac-stub/
--> Using cache 62d063a0b07081fa58c1307210002769d0b5dbb26aea2dcc4d348792efbc79d3
--> 62d063a0b070
[1/2] STEP 12/22: COPY harness/package.json ./harness/
--> Using cache dd0f04d1129719d61e5fa7ba5e929ef0b28c2528c1c48d24db6e5ac6f85c872e
--> dd0f04d11297
[1/2] STEP 13/22: COPY pi-fork/ ./pi-fork/
--> Using cache f824b923fc4910f8e3c7fb122e3d7b4fc95c7fb0a0f83d8265be27394a7778c0
--> f824b923fc49
[1/2] STEP 14/22: WORKDIR /app/pi-fork
--> Using cache 07d241370fb7203ae3dd080d2c09559adc16889704b06110db308780da3538dc
--> 07d241370fb7
[1/2] STEP 15/22: RUN npm ci
--> Using cache 6f1a1264a6f09d29ea36bd6799f0026b5b119caafcb0b5052378de99776d39c8
--> 6f1a1264a6f0
[1/2] STEP 16/22: WORKDIR /app
--> Using cache e805482ee8aa27490aabb041699f7959d6d07357f9b4de5fafa8ea7843917cdf
--> e805482ee8aa
[1/2] STEP 17/22: RUN pnpm install --frozen-lockfile
--> Using cache 5e7a0cf565d26c427b11df74cad391d26ab9748f05189d13c4872d328f4c5c38
--> 5e7a0cf565d2
[1/2] STEP 18/22: COPY packages/ ./packages/
--> Using cache 78326015eb130cb3134727077eecd36fbe61ff2b25f044eccb6d4f32f6627a83
--> 78326015eb13
[1/2] STEP 19/22: COPY harness/ ./harness/
--> Using cache 0b49329fd8b507313a994467825ffcfa8b64c1733d769a18e91629fe96cf3c7e
--> 0b49329fd8b5
[1/2] STEP 20/22: WORKDIR /app/pi-fork
--> Using cache be5fca023d79d5a6550845e13c2c16b81eb9481714dba29926ae71aa15647b00
--> be5fca023d79
[1/2] STEP 21/22: RUN npm run build -w packages/ai &&     npm run build -w packages/agent &&     npm run build -w packages/tui &&     npm run build -w packages/coding-agent

> @earendil-works/pi-ai@0.79.1 build
> npm run generate-models && npm run generate-image-models && tsgo -p tsconfig.build.json


> @earendil-works/pi-ai@0.79.1 generate-models
> node scripts/generate-models.ts

Fetching models from models.dev API...
Fetching models from NVIDIA NIM API...
Fetched 102 model IDs from NVIDIA NIM
Loaded 630 tool-capable models from models.dev
Fetching models from OpenRouter API...
Fetched 347 tool-capable models from OpenRouter
Fetching models from Vercel AI Gateway API...
Fetched 222 tool-capable models from Vercel AI Gateway
Generated src/models.generated.ts

Model Statistics:
  Total tool-capable models: 1261
  Reasoning-capable models: 984
  amazon-bedrock: 118 models
  anthropic: 13 models
  google: 22 models
  openai: 41 models
  groq: 6 models
  cerebras: 2 models
  cloudflare-workers-ai: 16 models
  cloudflare-ai-gateway: 57 models
  xai: 9 models
  zai: 5 models
  zai-coding-cn: 5 models
  mistral: 31 models
  huggingface: 66 models
  fireworks: 23 models
  nvidia: 31 models
  together: 19 models
  opencode: 61 models
  opencode-go: 19 models
  github-copilot: 33 models
  minimax: 3 models
  minimax-cn: 3 models
  kimi-coding: 4 models
  moonshotai: 10 models
  moonshotai-cn: 10 models
  xiaomi: 6 models
  xiaomi-token-plan-cn: 5 models
  xiaomi-token-plan-ams: 5 models
  xiaomi-token-plan-sgp: 5 models
  openrouter: 348 models
  vercel-ai-gateway: 222 models
  deepseek: 2 models
  ant-ling: 3 models
  openai-codex: 4 models
  google-vertex: 13 models
  azure-openai-responses: 41 models

> @earendil-works/pi-ai@0.79.1 generate-image-models
> node scripts/generate-image-models.ts

Fetching image models from OpenRouter API...
Fetched 45 image models from OpenRouter
Generated /app/pi-fork/packages/ai/src/image-models.generated.ts
src/models.generated.ts(10665,13): error TS2353: Object literal may only specify known properties, and '"supportsReasoningEffort"' does not exist in type 'OpenAIResponsesCompat'.
npm error Lifecycle script `build` failed with error:
npm error code 1
npm error path /app/pi-fork/packages/ai
npm error workspace @earendil-works/pi-ai@0.79.1
npm error location /app/pi-fork/packages/ai
npm error command failed
npm error command sh -c npm run generate-models && npm run generate-image-models && tsgo -p tsconfig.build.json
Error: building at STEP "RUN npm run build -w packages/ai &&     npm run build -w packages/agent &&     npm run build -w packages/tui &&     npm run build -w packages/coding-agent": while running runtime: exit status 1

Steps To Reproduce

docker build --load -t dev.local/serverless-harness:local .

Expected Behavior

I expect it to succeed.

Rossoctl Version

No response

Additional Context

This would close #154. However, this may have more to do with the pi-fork than this repo because the upstream API is incompatible now upon model fetch.

[pdettori]

This is resolved on main by #161 (merged 2026-08-24), the same fix that closed the duplicate #160.

Root cause: pi-fork/packages/ai's build script regenerated src/models.generated.ts from the live models.dev API at build time. models.dev added supportsReasoningEffort on a model mapped to the openai-responses API, which pi's pinned OpenAIResponsesCompat type didn't declare, so tsgo rejected the generated literal (error TS2353). Because generation hit a live, drifting API, the build broke retroactively with no code change (note the line number moved between reports: 10665 / 10611 / 10662).

Fix (#161): bumps pi-fork → rossoctl/pi@2c28be5, which (1) makes the build hermetic — npm run build now runs only tsgo against committed generated files, with codegen split into a deliberate generate step and a check:models-drift guard — and (2) widens OpenAIResponsesCompat to declare supportsReasoningEffort.

Verified on current main (submodule 2c28be50):

• npm run build -w packages/ai — the Dockerfile step (STEP 21) that failed here — exits 0 with no live model fetch.
• Full docker build --load -t dev.local/serverless-harness:local . completes successfully.

The "as a first-time quickstart I didn't expect to build the image" point was addressed separately in #164, which defaults to pulling the published image.

Please pull the latest main (git submodule update --init to sync pi-fork) and rebuild. Closing as fixed by #161.