Houston Cloud v0.5.37

Houston v0.5.37

Features

• feat(store): Agent Store redesign — shared web/app screens, one profile page, Drive-style sharing, listing editor, one-click hosted install (#1186)
• feat: personal-only AI accounts in team spaces (HOU-976) (#1187)
• feat: themed boot splash, Guide me in Settings, icon-only bell, Everyone-first scope, drop unread dot (#1175)
• feat(missions): user-driven Done, card confetti, mandatory follow-ups, archived UX (#1173)
• feat(web): responsive mobile layout for the shared shell (HOU-1014) (#1157)
• feat(integrations): add custom integration goes straight to the setup chat (#1171)
• feat(chat): custom integration actions brand the mission log (HOU-1049) (#1167)
• feat(chat): web + Python activity indicators in the process header (HOU-1048) (#1163)
• feat(skills): global Skills manager in the sidebar (HOU-792) (#1128)

Fixes

• fix(docker): copy patches/ into root-context images so pnpm install survives patchedDependencies (#1185)
• fix(integrations): setup research is invisible and fast - temp dir, batched fetches, add once (HOU-1083 follow-up) (#1184)
• fix(integrations): custom POSTs survive mixed undici stacks, manual add returns, deterministic discovery (HOU-1083) (#1182)
• fix(web): no gateway calls without a session; web account-switch resets caches (HOU-1014) (#1183)
• fix(chat): hard-wrapped URL labels render inline, not as a clipped pill (HOU-1071) (#1179)
• fix(app): offline gateway bursts surface one connectivity toast, not red bug toasts + Sentry noise (HOU-1085) (#1181)
• fix(providers): api-key providers reconnect via the key dialog; classify NVIDIA/Qwen auth rejections (HOU-1077) (#1178)
• fix(skills): share to workspace reaches the real workspace; local rows drop copy checkboxes (HOU-1027) (#1172)
• fix(web): cloud web boots to sign-in; skip pre-auth language/agreement gates (HOU-1014) (#1174)
• fix(auth): fall back to the device-code sign-in when the Codex loopback can't run (HOU-1063) (#1168)
• fix(integrations): custom action transport failures surface their real cause (HOU-1052) (#1162)
• fix(chat): keep the mission log live when tools follow streamed narration (HOU-1047) (#1161)
• fix(chat): agent-only markdown at a chat-tuned scale; user messages verbatim (HOU-1051) (#1164)
• fix(chat): suggestion pill click prefills the composer instead of sending (HOU-1050) (#1160)

CI & infra

• ci(web): preview web app bakes the staging gateway (#1177)
• ci(web): deploy the webapp from the cloud-v channel; staging tracks main (#1165)

Changes

• perf(e2e): parallelize the UI suite with per-worker fake hosts (HOU-1072) (#1166)

Before you upgrade

• Quit Houston before installing. macOS doesn't always replace a running app. If in doubt: rm -rf /Applications/Houston.app and drag the new copy in.

Full changelog: cloud-v0.5.35...cloud-v0.5.37

---

Linear — cloud-v0.5.37

(no project)

• HOU-1072 — Improve e2e test performance
• HOU-1022 — Design a beautiful loading screen
• HOU-1064 — Model selection is global per agent — changing the model in one mission changes it for all missions
• HOU-1065 — New agent defaults to Anthropic Sonnet even when Anthropic is not connected — first message always fails
• HOU-1073 — Managed cloud: Anthropic org credential killed by a second rotator — every NEW agent pod 401s while the UI says connected
• HOU-1080 — Every new mission opens in Ask First mode (drop Planner default + per-agent mode memory)
• HOU-1083 — Fix actions discovery when manually adding custom integrations
• HOU-1085 — Offline gateway bursts surface as red bug toasts + Sentry noise (list_project_files: Load failed)
• HOU-1071 — Fix link looking weird
• HOU-1014 — Verify app web is working

AI Models

• HOU-1077 — Nvidia and Qwen not working 🐛

Activity

• HOU-1043 — The Archived button should be primary and should be clear how to return to the main menu

Chat Experience

• HOU-1047 — When OpenAI does async stuff (e.g. connect to anakin.io) it shows the loading element in the chat instead of showing the activity, even if it is browsing the web, which SHOULD be something. The loading is ONLY if there's no connection with the model or regular wait times
• HOU-1051 — We shouldn't render the markdown from the user, only the one from the agent
• HOU-1050 — On the suggestions thing, on click it shouldn't send the message but put it as text in the chat input
• HOU-1049 — For the custom integrations also use this in the chat experience, maybe the name of the integration and a 🛠️ icon
• HOU-1048 — On the activity from the model it'll be great if we can show stuff like if the model is on the internet (they are curling stuff from there) and if the model for example is using python. Use an UI similar to the integrations, so find a logo and make look pretty. On the internet part, it'll be good if it is something like "(logo) Browsing the web - anakin.io"

Global Interfaces & Components

• HOU-1044 — The navbar should be reduced: compress the guide me and notifications to be icon only. Change the order of the teams members filter (default to everyone)

Integrations

• HOU-980 — Custom integrations should be working properly and should be fully tested
• HOU-1052 — Having an issue with anakin.io custom integration, this is what Houston agent said:

Multiplayer

• HOU-976 — Design: per-person vs team accounts for integrations and AI (both, with a resolution rule)

Onboarding

• HOU-1063 — OpenAI not connecting during onboarding 🐛

Skills

• HOU-1033 — [cloud] Gateway: /v1/orgs/:org/skills CRUD + Pg metadata + role enforcement
• HOU-1032 — [cloud] Pod-store: org-shared prefix read routes for engine pods
• HOU-1027 — Org skills store: org storage is the single source of truth
• HOU-792 — Custom skills are transferable to other agents (they are saved at the org level rather than the agent level)

WhatsApp draft

🚀 Nueva versión de Houston!
✅ When OpenAI does async stuff (e.g. connect to anakin.io) it shows the loading element in the chat instead of showing the activity, even if it is browsing the web, which SHOULD be something. The loading is ONLY if there's no connection with the model or regular wait times
✅ Improve e2e test performance
✅ The Archived button should be primary and should be clear how to return to the main menu
✅ Custom integrations should be working properly and should be fully tested
✅ Design a beautiful loading screen
✅ The navbar should be reduced: compress the guide me and notifications to be icon only. Change the order of the teams members filter (default to everyone)
✅ Model selection is global per agent — changing the model in one mission changes it for all missions
✅ New agent defaults to Anthropic Sonnet even when Anthropic is not connected — first message always fails
✅ Managed cloud: Anthropic org credential killed by a second rotator — every NEW agent pod 401s while the UI says connected
✅ Every new mission opens in Ask First mode (drop Planner default + per-agent mode memory)
✅ Nvidia and Qwen not working
✅ Fix actions discovery when manually adding custom integrations
✅ Offline gateway bursts surface as red bug toasts + Sentry noise (list_project_files: Load failed)
✅ Fix link looking weird
✅ OpenAI not connecting during onboarding
✅ Having an issue with anakin.io custom integration, this is what Houston agent said:
✅ We shouldn't render the markdown from the user, only the one from the agent
✅ On the suggestions thing, on click it shouldn't send the message but put it as text in the chat input
✅ For the custom integrations also use this in the chat experience, maybe the name of the integration and a 🛠️ icon
✅ On the activity from the model it'll be great if we can show stuff like if the model is on the internet (they are curling stuff from there) and if the model for example is using python. Use an UI similar to the integrations, so find a logo and make look pretty. On the internet part, it'll be good if it is something like "(logo) Browsing the web - anakin.io"
✅ [cloud] Gateway: /v1/orgs/:org/skills CRUD + Pg metadata + role enforcement
✅ [cloud] Pod-store: org-shared prefix read routes for engine pods
✅ Org skills store: org storage is the single source of truth
✅ Verify app web is working
✅ Design: per-person vs team accounts for integrations and AI (both, with a resolution rule)
✅ Custom skills are transferable to other agents (they are saved at the org level rather than the agent level)
Gracias a todos los que reportaron 🙌

Notify reporters (then clear the Notify pending label):

• HOU-1077 → ⚠️ no phone on issue
• HOU-1063 → ⚠️ no phone on issue