Houston Cloud v0.5.29

Houston v0.5.29

Features

• feat(landing): @mentions, Features section, auto-rotating switchers
• feat(landing): free tier for three, hero share preview, full cleanup
• feat(landing): v4 final round — cloud hero, category copy, nav rework
• feat(prototype): sky-2 hero + cool blue palette
• feat(prototype): round 9 — expert-per-job column, richer agent views, FAQ accordions
• feat(prototype): round 8 — agent column switcher + feature views, inline
• feat(prototype): round 7 — screen-per-section polish, inline
• feat(prototype): round 6 — clear statements + app screens, built inline
• feat: broken connections recover in place, not in a special section (HOU-847 follow-up)
• feat(prototype): occupied-board hero — the multiplayer fold
• feat(prototype): first-principles refactor — ivory, visual-first, 245 words
• feat: merge the connect flow into the row as one card (HOU-847 follow-up)
• feat(prototype): multiplayer-led story on Houston monochrome (HOU-918 round 2)
• feat(prototype): Mercury-look skin for the landing (HOU-918, do not deploy)
• feat: polish the integrations connect flow (HOU-847)
• feat(models): add Claude Opus 5 (HOU-912) (#1090)
• feat(app): instant setup-mission hello while the pod warms (HOU-867) (#1081)
• feat: landing polish — exact-app preview, compact how-it-works, 8-agent grid
• feat: richer skills.sh preview modal with category, apps, and instructions (#1092)
• feat(agents): remove the Connect (public API) section from Agent Settings (HOU-806) (#1075)
• feat: update pi (Earendil) dependencies 0.80.6 → 0.82.0 (HOU-891) (#1083)
• feat: show a skill's connected apps across the skill surfaces (#1091)
• feat: make notifications discoverable — contextual ask, honest settings, catch-net (HOU-879)
• feat: landing v3 — original sections, copy, and animations on the light theme
• feat: render skill icons with the flat 2D Fluent emoji variant (#1087)
• feat: astronaut one-liners on the thinking indicator (HOU-910)
• feat: human labels for every Houston tool in the mission log (HOU-888)
• feat: icon-first mission status, no prefix (HOU-888)
• feat: landing v2 — monochrome editorial restyle with true-to-app preview
• feat(files): upload complete folders from the Files tab (HOU-889) (#1077)
• feat(ci): sign Windows MSI via Azure Artifact Signing to clear SmartScreen (#1082)
• feat: branded mission status while an app action runs (HOU-888)
• feat: rebuild landing page — Night Launch design on a single token layer
• feat: one card system — branded ask_user replaces the approval gate (HOU-885)
• feat(teams): teammate names and avatars from the gateway profile source
• feat: agent design process — DESIGN.md + frontend-design/design-review skills
• feat: visual regression suite + design-token hex cleanup
• feat: ask_user defaults to options; honest optionless placeholder (HOU-880)
• feat: forward composerOverrideMode through the board panel (HOU-870)
• feat: interaction cards replace the composer, free-text escape on every step (HOU-870)
• feat: integration approvals become a plain confirmation card (HOU-869)
• feat: make display name optional in the creator profile forms (#1064)
• feat(migration): emoji space-invaders + progress screen polish (#1063)
• feat: rename Coworker mode to Ask first
• feat: run read-only integration actions without the approval card
• feat(chat): upload whole folders as chat attachments (HOU-808) (#1045)
• feat: label Linear issues at the cut, not just at publish (#1049)
• feat: presence-style status dot on catalog rows (#1040)
• feat: sign-in referral panel reads Coming soon; localize panel strings (#1039)
• feat: Linear release stamp on cloud train publish (#1038)
• feat(store): starter-agent publish pipeline, admin grant UI, docs and i18n cleanup
• feat(integrations): turn the catalog spotlight into a "Most used" section
• feat(analytics): stamp the display name as a person property on sign-in
• feat(analytics): emit user_signed_up on first sign-in of a new GCIP account
• feat(providers): 2026-07 catalog QA sweep + Gemini 3.6 Flash / 3.5 Flash-Lite
• feat(analytics): emit $ai_generation per model turn to light up the AI Usage dashboard
• feat(routines): chat-first redesign of the Routines tab
• feat(catalog): add Kimi K3 (kimi-coding) by backporting the pi-ai 0.80.9 entry
• feat(store): creator profiles - in-app editor, @handle pages, analytics, verified badges
• feat(app): unified catalog search across integrations, AI hub, and skills
• feat(ui): two-section catalog shell with unified search grammar
• feat(providers): hide regional duplicate deployments from the catalog
• feat(store): one-click Open in Houston install from the website
• feat(analytics): instrument usage, permissions, org, AI hub, skills, cloud migration
• feat(analytics): instrument feature-adoption gaps + model/provider on sends
• feat(sentry): tag client events with their deployment (#991)
• feat(dev): add pnpm dev:staging to run the desktop against remote staging (#994)
• feat(app): send app-version header to the gateway + blocking required-update screen (#985)
• feat(host): build identity on /v1/version + pod-level /activity busy probe (#984)
• feat(chat): apply AI Mode changes to the running turn, like Claude Code's shift+tab
• feat(files): transparent files panel + move-conflict dialog (#966)
• feat(store): app parity wave - categories, integration filter, reporting, owner dashboard
• feat(store): extract @houston/agentstore-client SDK and migrate both frontends
• feat(teams): agent-centric Permissions, on-agent mount, and permission-aware agents (#954)
• feat(teams): admin Permissions hub — People access lens + blocked-state deep links (#948)
• feat(usage): split the Usage page into Compute usage / Model usage panes
• feat(integrations): review and revoke always-allowed actions per agent (#942)
• feat(integrations): surface connected-but-ungranted apps on the agent tab (#939)
• feat(files): redesign the Files tab as a Drive-style card grid (#960)
• feat(analytics): enable session recordings + heatmaps (masked)
• feat(chat): add conversation map
• feat(onboarding): email steps - floating brand marks, one clear final action
• feat(onboarding): preselect the OS locale on the language gate
• feat(chat): toast when a mode change lands on the next message
• feat(usage): show time worked instead of engine up-time + chart legibility (#936)
• feat(usage): split accounts into AI subscriptions and AI per token sections
• feat(sentry): debuggable engine events — synthetic stacks, version tag, severity correctness (#937)
• feat(ios): move auth to GCIP with Apple, Google, Microsoft, and email-code sign-in (#938)
• feat(secrets): use gateway custody for managed agents (#899)
• feat(usage): show how long agents ran (compute usage, hosted cloud) (#925)
• feat(models): curate provider model lists via one shared visibility table (#928)
• feat(create-agent): model picker in the Create-with-AI step (#927)
• feat(usage): meter token spend locally for API-key providers with no usage API
• feat(chat): preview + download files the agent created from chat (#923)
• feat(onboarding): larger Gmail/Outlook rows on the connect-email step
• feat(onboarding): centered pill-grid segment screen with skip + PostHog person property
• feat(analytics): enable rage + dead click capture, fully masked
• feat(engine): Sentry crash reporting for the TS engine (pods + desktop sidecar) (#907)
• feat(onboarding): add reset segment setting
• feat(onboarding): add segmentation screen
• feat(website): unlisted bootcamp landing page at /bootcamp/ (#906)
• feat: Admin > Agents drill-in to manage one agent's access in place
• feat: visible workspace policy everywhere and a settings-style Admin page
• feat: move org allowlists to an Admin page, keep Integrations personal
• feat(dev): one pnpm dev — full local stack, multiplayer included (#900)
• feat(migration): reconnect checklist from the legacy Composio consumer account (#895)
• feat(local-model): share a tunnelled local model with your team (#860)
• feat(auth): provider pill row + six-box pin input on the login screen
• feat(design): visible sidebar selected state via sidebar-active token
• feat(board): say Archive/Archived instead of Complete/Completed
• feat: per-account provider usage page (top-level Usage view) (#890)

Fixes

• fix(website): mobile excellence pass
• fix(landing): footer + resources cleanup
• fix(landing): drop the per-OS line from the pricing card
• fix(prototype): let the hero sky breathe
• fix(prototype): hero readability + transparent nav
• fix(prototype): hero headline joins the content grid
• fix: restore Windows download path for non-Windows visitors
• fix(runtime): resume a renamed agent's conversations — relocate the transcript, retry fresh as last resort (HOU-892) (#1094)
• fix: harden the bug-report path and the Settings workspace gate
• fix: move how-it-works heading into the left column
• fix(HOU-892): routines incident — stale Claude-credential overwrite guard + idempotent Composio disconnect (#1079)
• fix(runtime): paragraph break between a turn's streamed text blocks (HOU-857) (#1076)
• fix(board): never paint the previous agent's missions on switch (HOU-858) (#1078)
• fix(host): accept a nonce-matched re-send of the running turn instead of 409 (HOU-807) (#1085)
• fix(teams): first prod test round - identity reset, in-place switch, per-space provider state
• fix(release): stamp the whole train in Linear; clean up superseded drafts (#1074)
• fix(teams): bridge team spaces into the switcher; invite follow-through; spaces e2e
• fix(teams): personal space never offers a broken invite flow
• fix(auth): register the houston:// scheme at runtime on Windows
• fix: surface the real cause when a creator profile save or photo upload fails (#1062)
• fix(i18n): translate the pre-auth sign-in screen (es/pt)
• fix(analytics): let session replay start by not disabling flags
• fix(providers): report a provider connected only when it can actually serve a turn (#1054)
• fix(credentials): make the desktop reconcile a fill-only push so it cannot revoke the token family (HOU-855) (#1056)
• fix(runtime): show the connect card in Autopilot mode (HOU-853) (#1055)
• fix(chat): flush a stranded reconnect auto-continue with a history-probe watchdog (HOU-849) (#1052)
• fix(integrations): serve the secure credential card through the per-agent surface so managed-cloud saves stop 404ing (HOU-823) (#1053)
• fix(app): accept the Claude sign-in code when the browser hand-off is blocked (HOU-839) (#1043)
• fix: keep the open chat when clicking outside the board panel (#1048)
• fix(teams): resume abandoned agent moves so a wedged move can't brick the agent (HOU-817) (#1044)
• fix(ci): green-main guard must ignore the cut's own failed runs
• fix(host): stop exhausting the Linux inotify watch budget (HOU-841) (#1042)
• fix(ci): surface GitHub's rejection message in the cut token probe
• fix(ci): tokenless checkout + API-validated PAT for the daily cut
• fix: show the full email on the sign-in continue button (#1041)
• fix: never sign users out on secure-storage read faults (#1037)
• fix(integrations): platform-correct custom-secret file permissions
• fix(integrations): never dead-end the secure credential save on schemeless specs
• fix(host): quiesce of an absent runtime is a no-op, not a launcher sleep error
• fix(analytics): identify the person before emitting user_signed_up
• fix(providers): route Copilot Business connects through github.com
• fix(providers): drop the two Cloudflare providers from the catalog
• fix(runtime): classify no-credit provider errors as quota_exhausted
• fix(host): quiesce the agent runtime before a rename so the old name cannot resurrect
• fix(host): serve the Rust-era agent color so migrated agents keep their colors
• fix(integrations): pin Composio tool version to latest on search + execute
• fix(chat): block image attachments when the active model has no vision
• fix(host): no-auth Composio toolkits are not connectable apps (#1012)
• fix(chat): render URL-labeled links inline instead of a clipped black pill
• fix(app): repair Linux AppImage sign-in end to end (#1019)
• fix(providers): friendly toast for a GitHub account without Copilot access (#1017)
• fix(providers): verify Google keys against the models list and surface typed connect errors (#1016)
• fix(providers): repair the Windows shell env for Claude sign-in and turns (#1014)
• fix(app): guard agent-JSON reads against wrong top-level container shape (#1013)
• fix(website): download buttons serve the newest cloud prerelease (#1009)
• fix(routines): adapt intake connect card to per-slug connect-flow states
• fix(onboarding): migration-flow polish, durable onboarding-completed signal, deleted-account sign-out (#999)
• fix(store-sync): skip an over-cap file instead of letting it block every sync pass (#989)
• fix(runtime): size Claude context fill from per-request usage, not the turn aggregate (#988)
• fix(sentry): stop expected operational failures from spamming error events (#986)
• fix(runtime-client): retry transient object-store failures instead of failing hydrate/sync (#987)
• fix(runtime-client): tolerate files vanishing mid-walk in syncBack (#983)
• fix(host): cover all delivery paths of the recursive-watcher ENOENT race (#981)
• fix(app): reconnect card for a disconnected local model + unknown-state polish (#980)
• fix(local-model): connect directly when the deployment has no tunnel relay (#978)
• fix(chat): unwedge the reconnect auto-continue from the send queue (#979)
• fix(host): disconnect a terminally-rejected credential instead of retrying its dead refresh token forever (#977)
• fix(skills): require a description before creating a skill from scratch (#976)
• fix(app): keep the Codex sign-in relay off for a loopback hosted gateway (#955)
• fix(website): unblock download gate and de-jank modal open (#975)
• fix(providers): report unknown, not signed-out, when the engine is unreachable (#973)
• fix(host): demote the Linux recursive-watcher ENOENT race to a warning (#970)
• fix: stop a double-fired first send from losing a mission (#969)
• fix(usage): only show the user's own agents, count messages not tasks (#967)
• fix(onboarding): skip only on failure; the sent payoff says check your inbox
• fix(onboarding): email mission runs in Autopilot; skip waits for the agent's reply
• fix(providers): fail Codex sign-in port conflicts fast, loud, and localized (#945)
• fix(e2e): conversation map spec earns its 3-moment minimum on the fake host
• fix(agentstore): pin folder-entry timestamps — reproducible skill zips (#961)
• fix(sentry): count orgs as users-affected, trim the runtime capture frame (#957)
• fix(usage): hide system/ghost agents from the Time worked list (#956)
• fix(analytics): admit detected_locale through the branch cleanProps filter
• fix(chat): conversation map floats over the transcript + clean marker previews
• fix(onboarding): ask the work-segment question at most once per user
• fix(web): provider connect no longer reports success it did not earn
• fix(auth): desktop Apple sign-in returns via gateway bridge + houston:// deep link
• fix(app): paint cached mission cards on cold open while the pod wakes (#950)
• fix(sentry): attach the synthetic stack as a thread, keep the message as the issue title (#949)
• fix(design): floating surfaces are solid everywhere, no glass modals
• fix(chat): approval card drops the technical param rows
• fix(host): satisfy noUncheckedIndexedAccess in the provider-usage route test
• fix(usage): Copilot usage no longer reads "sign in again" on a healthy connection
• fix(onboarding): require initial email action
• fix(chat): pin the agent's configured model on setup-chat kickoffs
• fix(runtime): expire abandoned OAuth logins so they release the callback port (#932)
• fix(settings): remove onboarding segment reset from production settings
• fix(design): solid dialog + create-agent cards, no see-through surfaces
• fix(e2e): scope usage-compute agent rows to the Running time section
• fix(design): opaque floating surfaces on desktop (WebView2 backdrop-filter)
• fix(chat): make pasted links readable inside the user message bubble
• fix(usage): honest OpenRouter balance + "not metered yet" copy
• fix(routines): stop cron routines vanishing after an edit (#924)
• fix(fake-host): split the setup-runtime connect surface into gate + first-run slots
• fix(e2e): onboarding connect spec survives the connected setup-runtime
• fix(providers): live-verify a pasted API key before storing it
• fix(web): make the engine gate reachability-only; onboarding owns connect
• fix(chat): make the in-chat integration approval card work on hosted cloud + stop the footer overflowing (#919)
• fix(fake-host): serve the /setup-runtime connect surface the WebApp gate probes
• fix(web): point the standalone connect gate at /setup-runtime
• fix(e2e): teach the fake host the /setup-runtime connect surface
• fix(web): probe the host's /setup-runtime connect surface, not a flat /auth/status
• fix(web): point the connect gate at the host's /setup-runtime surface
• fix(providers): curate key-dashboard URLs for the remaining api-key providers
• fix(providers): connect any pi api-key provider from the UI
• fix(integrations): never mint a bare custom OAuth auth config (Meta Ads)
• fix(onboarding): email test-send card shows only the option; Omitir actually skips
• fix(app): paint sidebar needs-you badges through a pod cold start (#914)
• fix(onboarding): drop the "Another provider" row from the connect-email step
• fix: cloud-migration wizard recovery — keep the Settings retry after partial runs, escape hatch on errors (#908)
• fix(anthropic): repair OAuth connect + credential lifecycle end to end (#910)
• fix(app): keep cached mission cards painted while an asleep pod wakes (#911)
• fix(ai-hub): stop curating OAuth providers' model lists — hub now matches the chat picker (#909)
• fix(onboarding): adapt segmentation screen to current main after rebase
• fix: cloud-migration backup fails on Windows with 'Acceso denegado (os error 5)' (#905)
• fix(sidebar): stop agent list overflowing onto the user footer in short windows (#898)
• fix(release): preserve Bun sidecar in Linux AppImage (#897)
• fix(release): give vite build 4GB Node heap on low-RAM runners (#896)
• fix(chat): stop hiding a reply whose text contains an auth code (HOU-734) (#885)
• fix(migration): skip the empty Reconnect-your-apps step (#893)
• fix(ui): keep the open chat panel mounted while its card is transiently absent (#891)

Docs

• docs: add the app node:test suite to the test-commands table
• docs: KB reflects composer-replacing interaction stepper (HOU-870)
• docs(kb): drop the resolved sign-in-screen i18n follow-up note
• docs: record the catalog redesign in the knowledge base and inventory
• docs(kb): document agentstore-client SDK, store parity features, deep-link install
• docs(website): point developer docs at the real gateway, centralize the URL (#993)
• docs(kb): drop retired theagentlibrary note from agent-store KB

CI & infra

• ci: retrigger
• ci: retrigger
• ci: raise the web job timeout to 30m — the e2e suite outgrew 20m
• ci(train): publishing the cloud draft ships the whole train
• ci(images): publish the served /v1/catalog as an engine-pod-catalog artifact
• ci(release): cut the cloud release automatically each weekday morning
• ci(release): build a staging-gateway QA DMG for every cloud release
• ci(website): pass POSTHOG_KEY to the site build and fail loudly when absent
• ci(images): build engine-pod/agentstore on tag push, notify prod sepa… (#992)

Changes

• test(runtime): pin the mid-turn Mode-pill hot-swap on the Claude SDK backend (HOU-776) (#1080)
• test: notification body-key cases follow the branded-question model (HOU-885)
• test: prove the Always-allow circle end-to-end through the dispatch route
• perf(chat): open chats on the transcript tail + scroll-up lazy-load (HOU-819) (#1051)
• style(chat): drop the amber triangle chip from the attachment-rejection dialog
• test(routines): adapt stale e2e specs to the chat-first Routines redesign (#1010)
• refactor(files): flatten the Files tab onto the canvas (#1007)
• test(e2e): update catalog specs for the unified catalog redesign
• test(onboarding): skip-route assertion follows the failure-gated shouldOfferSkip
• refactor(settings): drop the Connected accounts shortcut row (#947)
• refactor(integrations): fold Settings > Connected accounts into the Integrations page (#941)
• test(onboarding): update language-gate e2e to the preselect + Continue contract
• test(web): approval e2e pins the card without param rows
• test(onboarding): select required segment in e2e
• refactor(usage): drop the billing sections' subtitles
• test(fake-host): serve the pre-agent /setup-runtime/* connect surface
• test(e2e): align onboarding specs with the segment gate and the welcome-less flow
• perf(host): parallelize store hydration — cuts managed-pod wake by ~10s (#901)

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: v0.5.6...cloud-v0.5.29

---

Linear — cloud-v0.5.29

(no project)

• HOU-900 — Landing v2: monochrome editorial restyle (no space bg, no orange, real-app mockup)
• HOU-908 — Landing v3: original sections/copy/animations on the light monochrome theme
• HOU-913 — Landing polish: exact-app preview, compact how-it-works layout, 8 agents + store tile
• HOU-919 — Landing: no Windows download path from non-Windows browsers (per-OS triggers lost in v3)
• HOU-918 — PROTOTYPE (do not merge): Mercury-look landing for Mercury CTO review
• HOU-933 — Landing: free tier up to 3 members + OG hero preview + Mercury/dead-code cleanup
• HOU-940 — Landing: @mentions in chats, Features section (2 parts), auto-rotating switchers
• HOU-934 — Website: migrate Resources subpages (guides, vision, changelog) to the v4 design system + shared navbar
• HOU-935 — Website: mobile excellence pass across landing + migrated subpages

AI Models

• HOU-891 — Update Pi
• HOU-912 — Add support for Opus 5
• HOU-776 — Anthropic: Models should be able to change from Coworker, Planner, and Autopilot modes while they are responding (hot-swap) like it works in Pi

Agents API

• HOU-806 — Remove this from the agents settings screen, it should be only in Routines for now. Just delete the UI.

Backend

• HOU-897 — Alert when a woken engine hydrates 0 objects for an agent the gateway knows
• HOU-896 — Roll guard: never recycle an engine pod whose data has not reached the pod-store
• HOU-895 — Gateway self-heal: re-seed known agents into a blank engine instead of eternal 404s

Chat Experience

• HOU-888 — We have a "Mission in progress" that has a ":" and that changes with the detail of what we're doing. When we're using an integration, we probably have also the logo. It'll be amazing to show the logo and the name of the integration instead of doing the generic integration execute
• HOU-910 — Astronaut loading phrases: the pulsing helmet cycles ~50 fun space one-liners
• HOU-879 — Desktop notifications when a mission needs you or finishes
• HOU-939 — Context window of a chat is not correct. E.g. Opus 4.8 should be 1M, we're allowing 250K
• HOU-937 — Agent responses are appearing in one long paragraph instead of clean sentences
• HOU-936 — Uploaded files are not "landing" in the agent's workspace
• HOU-857 — Fix response streaming

Files

• HOU-889 — On the files tab, users should be able to upload complete folders

Integrations

• HOU-847 — While integrations are being connected they show an ugly experience on the integrations tabs, improve it

Multiplayer

• HOU-903 — Sign-out does not reset in-memory state: next account renders the previous account's world
• HOU-904 — Entering a team on a billing-off gateway throws the red bug toast (billing 503 not degraded)
• HOU-905 — Create-team from Settings leaves the user in Settings: no navigation into the new team
• HOU-906 — Provider connections from the personal space render inside a team after a space switch
• HOU-907 — Space switch shows a blocking loading screen: should be a seamless in-place transition

Performance

• HOU-858 — Fix chat loading
• HOU-867 — Fix agent creation

Scheduled

• HOU-892 — Review that routines are working properly

Skills

• HOU-793 — Skills icons should be 2D
• HOU-794 — Nice to show the apps connected to that skill (for custom Houston skills)
• HOU-795 — Add more info from skill.sh into the skill information modal

WhatsApp draft

🚀 Nueva versión de Houston!
✅ On the files tab, users should be able to upload complete folders
✅ Landing v2: monochrome editorial restyle (no space bg, no orange, real-app mockup)
✅ Skills icons should be 2D
✅ Sign-out does not reset in-memory state: next account renders the previous account's world
✅ Entering a team on a billing-off gateway throws the red bug toast (billing 503 not degraded)
✅ Create-team from Settings leaves the user in Settings: no navigation into the new team
✅ Provider connections from the personal space render inside a team after a space switch
✅ Space switch shows a blocking loading screen: should be a seamless in-place transition
✅ Landing v3: original sections/copy/animations on the light monochrome theme
✅ Nice to show the apps connected to that skill (for custom Houston skills)
✅ Update Pi
✅ Add more info from skill.sh into the skill information modal
✅ Fix chat loading
✅ Landing polish: exact-app preview, compact how-it-works layout, 8 agents + store tile
✅ We have a "Mission in progress" that has a ":" and that changes with the detail of what we're doing. When we're using an integration, we probably have also the logo. It'll be amazing to show the logo and the name of the integration instead of doing the generic integration execute
✅ Astronaut loading phrases: the pulsing helmet cycles ~50 fun space one-liners
✅ Desktop notifications when a mission needs you or finishes
✅ While integrations are being connected they show an ugly experience on the integrations tabs, improve it
✅ Landing: no Windows download path from non-Windows browsers (per-OS triggers lost in v3)
✅ PROTOTYPE (do not merge): Mercury-look landing for Mercury CTO review
✅ Landing: free tier up to 3 members + OG hero preview + Mercury/dead-code cleanup
✅ Landing: @mentions in chats, Features section (2 parts), auto-rotating switchers
✅ Website: migrate Resources subpages (guides, vision, changelog) to the v4 design system + shared navbar
✅ Website: mobile excellence pass across landing + migrated subpages
✅ Context window of a chat is not correct. E.g. Opus 4.8 should be 1M, we're allowing 250K
✅ Agent responses are appearing in one long paragraph instead of clean sentences
✅ Uploaded files are not "landing" in the agent's workspace
✅ Add support for Opus 5
✅ Alert when a woken engine hydrates 0 objects for an agent the gateway knows
✅ Roll guard: never recycle an engine pod whose data has not reached the pod-store
✅ Gateway self-heal: re-seed known agents into a blank engine instead of eternal 404s
✅ Review that routines are working properly
✅ Fix agent creation
✅ Fix response streaming
✅ Remove this from the agents settings screen, it should be only in Routines for now. Just delete the UI.
✅ Anthropic: Models should be able to change from Coworker, Planner, and Autopilot modes while they are responding (hot-swap) like it works in Pi
Gracias a todos los que reportaron 🙌