upstream-sync: merge simstudioai/sim main into feat/github-merge-agent (2026-08-04) - #680
Closed
utcarshsrivastava-collab wants to merge 534 commits into
Closed
upstream-sync: merge simstudioai/sim main into feat/github-merge-agent (2026-08-04)#680utcarshsrivastava-collab wants to merge 534 commits into
utcarshsrivastava-collab wants to merge 534 commits into
Conversation
…se (simstudioai#5906) * fix(landing): restore single-paragraph hero copy on home and enterprise * fix(landing): tighten enterprise hero description * revert(landing): restore original home and enterprise hero headings * improvement(landing): simplify enterprise hero description * improvement(landing): restore building-and-managing homepage headline
… pricing (simstudioai#5913) * improvement(library): add citations and internal links, correct stale pricing The library had zero third-party citations across 16 posts (~51k words) and averaged 1.2 internal links per post, with six posts at zero. Auditing every dollar figure against the vendor's own pricing page while adding the citations turned up several claims that had gone stale, plus one product that is being shut down. Corrections, each verified against the vendor's page on 2026-07-23: - Relay.app is winding down (signups closed 2026-07-16, free accounts end 2026-08-15, paid 2026-09-14). It was recommended as the human-in-the-loop pick in best-zapier-alternatives; that recommendation now points to a platform you can still sign up for - Make Core is $12/mo billed monthly, not $10.59; annual saves ~15%. One post also described $12 as the annual rate - Pabbly Connect lifetime starts at $349, not $249, and the Standard/Pro/ Ultimate tier structure quoted no longer exists - Workato publishes no pricing at all, so the "~$1,000/month" figure is replaced with the fact that every deal is quoted through sales - Dify is at ~149k GitHub stars, not 131k - n8n cloud Starter is EUR 20/mo billed annually for 2,500 executions Verified-correct figures were left alone and given a source link: Zapier $29.99/mo monthly for 750 tasks, Activepieces 10 free flows then $5/flow/mo, Power Automate $15/user/mo, Lindy $49.99/mo, and the Grand View Research RPA market figures. Also: 59 outbound citations and 3-6 internal links per post (zero posts left without internal links), and MDX external links now carry target=_blank plus rel=noopener noreferrer, which the landing SEO/GEO rule requires but the MDX anchor was not applying. * fix(content): treat only non-Sim hosts as external links in MDX The first pass classified any http(s) href as external, so the 33 absolute Sim URLs already in content (sim.ai, sim.ai/slack, www.sim.ai/blog/*, and 14 docs.sim.ai pages) would have opened in a new tab with rel=noopener, which is wrong for first-party links and contradicted the comment above the check. Classification now compares the hostname against the apex derived from SITE_URL, so the apex and any Sim subdomain stay same-tab. SITE_URL is used rather than getBaseDomain() so a post renders identically in dev, preview, and production instead of varying with NEXT_PUBLIC_APP_URL. The leading dot in the suffix check keeps lookalikes such as evil-sim.ai external.
…oard (simstudioai#5907) * fix(helm): correct chart docs, examples, and dead config across the board Audit-driven accuracy pass over the chart's entire documentation surface, verified by rendering every example against the templates: - migrations run as an init container on the app pod, not a Job — fix the README component list, troubleshooting commands, and sim-helm skill refs; drop the dead migrations-job NetworkPolicy ingress rule - referenced-but-never-created resources: document the GKE ManagedCertificate creation (values-gcp), comment out the key-file Secret mount that stuck all pods in ContainerCreating (values-gcp), enable certManager for the postgres TLS issuerRef (values-production), add the cert-manager cluster-issuer annotation nginx needs (values-azure) - values-external-db: networkPolicy.egress is a list, not a map (the map rendered an invalid manifest); fill schema-failing placeholder host/username - realtime >1 replica requires REDIS_URL (Socket.IO Redis adapter) — default examples to 1 replica with the scaling note, and warn where autoscaling HPAs override replicaCount - pod anti-affinity selectors matched nothing (simstudio vs sim name label) - kubernetes.mdx: install commands were missing required CRON_SECRET and postgresql password (failed at template time), wrong deployment name in port-forward, stale version requirements, unsupported key-remapping claim - remove unimplemented app.secrets.existingSecret.keys from values + schema; fix README PDB default, cronjob list, /metrics caveat, NOTES secret count, Azure-only StorageClass in generic examples, dead SOCKET_SERVER_URL and GOOGLE_CLOUD_* env, ESO apiVersion mismatch, and skill-reference drift - bump chart to 1.0.1 * fix(helm): review round 1 — scoped example egress, in-tab secret note, copilot Job wording - external-db example egress scopes to a placeholder database CIDR instead of to: [] (which allowed every destination on 5432, defeating the isolation the example teaches) - kubernetes.mdx cloud tabs state explicitly that they reuse the variables generated in the Installation block - Copilot migrations really do run as a Helm-hook Job — restore Job wording there (only the app migrations are an init container) * fix(helm): template-sweep fixes — telemetry validity, ESO rollout checksums, dead passwordKey knob - telemetry: memory_limiter gets the required check_interval (collector failed startup validation whenever telemetry.enabled=true); the jaeger exporter was removed from collector-contrib in v0.86 — export to Jaeger via its native OTLP endpoint instead (otlp/jaeger, default port 4317) - app/realtime rollout checksums now hash the ExternalSecret manifest too, mirroring the copilot pattern — with ESO enabled the inline Secret renders empty, so remoteRefs changes never rolled the pods - remove the unimplemented existingSecret.passwordKey knob (values, schema, README, dead helpers): nothing consumed it, and a non-default value silently produced a DATABASE_URL with an unexpandable placeholder; secrets must use the standard POSTGRES_PASSWORD / EXTERNAL_DB_PASSWORD keys - drop the orphaned sim.migrations.labels helper (its only consumer was the dead NetworkPolicy rule removed earlier) - helm test pod image resolves through sim.image so global.imageRegistry mirroring applies; NetworkPolicy realtime-ingress comment reflects actual traffic direction; smoke unittest suite loads the newly referenced external-secret template * chore(helm): bump chart to 1.1.0 with upgrade notes Removing (inert) documented values keys and changing the rollout-checksum inputs is a values-surface change — per SemVer chart conventions that is more than a patch. Adds an Upgrading section documenting the one-time pod roll, the removed no-op keys, and the Jaeger-over-OTLP change. * fix(helm): review round — external-db NP opt-in with real-CIDR-first flow, prod Jaeger OTLP endpoint - external-db example ships networkPolicy disabled so a verbatim install always reaches the database; the scoped egress rule stays as the documented opt-in (set your CIDR first, then enable) - values-production still pointed telemetry.jaeger at the legacy 14250 collector port — now Jaeger's OTLP gRPC endpoint to match the otlp/jaeger exporter * feat(helm): autoscaling.realtime.enabled toggle so examples can scale the app without unsafe realtime replicas Cursor correctly flagged that comment-level warnings didn't stop a verbatim production/external-db install from running the realtime HPA at minReplicas 2 without REDIS_URL (silent cross-pod event loss). Adds an opt-out toggle (default true — existing deployments unchanged): the realtime HPA renders only when autoscaling.realtime.enabled, and the realtime Deployment keeps spec.replicas under its control when the HPA is excluded. The three autoscaling examples set it false with the Redis rationale; README and upgrade notes document the toggle. * fix(helm): review round — whitelabeled realtime HPA opt-out, external-db isolation on with required CIDR in install flow - values-whitelabeled now actually sets autoscaling.realtime.enabled: false (the earlier batch aborted before reaching this file — Cursor caught it) - external-db keeps networkPolicy enabled (no isolation regression); the DB egress CIDR is marked REQUIRED and wired into both documented install commands via --set, so the copy-paste flow sets the real subnet in the same breath as the DB host * fix(helm): use a syntactically valid example CIDR in the external-db install commands An unreplaced <YOUR_DB_CIDR> literal fails Kubernetes CIDR validation and aborts the install; every sibling placeholder in the same command (host, username) installs fine and simply doesn't connect until replaced. The CIDR now behaves the same way: valid example value (10.20.0.0/24), explicitly marked as the operator's database subnet in both commands and the values comment. * fix(helm): remove text after continuation backslashes in external-db install commands A trailing comment after the line-continuation backslash (and equally a comment line spliced mid-command) breaks the shell command when copied — the remaining --set overrides run as separate commands and required-value validation fails. Both documented commands now reconstruct to bash-clean multi-line invocations (verified with bash -n); the CIDR guidance lives in the networkPolicy section comment. * fix(docs): cloud-tab installs are alternatives via helm upgrade --install Following Installation and then a cloud tab ran helm install twice for the same release and failed on the second. The tabs now state they replace the generic install and use helm upgrade --install, which is idempotent and also converts an existing generic install to the cloud values. * fix(docs): honest conversion caveat for cloud-tab upgrades over an existing install The cloud values rename the bundled Postgres database to simstudio, which Postgres only applies at first initialization — an in-place conversion of a generic install would point DATABASE_URL at a nonexistent database. Document the two safe paths: keep the original name via --set, or uninstall + delete PVCs and install fresh. * fix(helm): external-db header install command declares its secrets and includes CRON_SECRET The primary documented command failed the chart's required-value validation (missing app.env.CRON_SECRET with cronjobs default-on) and referenced an undeclared DB_PASSWORD. It now shows the export lines for every variable it uses and sets CRON_SECRET; both commands verified with bash -n. * chore(helm): restore values.schema.json formatting — surgical deletions only The earlier programmatic edit reformatted the whole file (~390 lines of whitespace churn hiding the 12 real deleted lines). Re-applied the removal of the dead keys/passwordKey properties as text-level deletions preserving the original style. * fix(helm+docs): explicit secret exports in external-db header; conversion must reuse original secrets - all five export lines are written out (three were only named in a trailing comment, so a verbatim copy passed empty required values) - the cloud-conversion caveat now leads with reusing the original secret values (helm get values) — a regenerated ENCRYPTION_KEY makes previously encrypted credentials undecryptable
* feat(agent-stream): add agent-events thinking/tool streaming for chat and canvas
Ship the agent-events-v1 protocol with provider tool loops, dual-gated chat thinking, DeepSeek/Groq/OpenAI reasoning wiring, and ChatGPT-like thinking chrome.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(agent-stream): clear stuck streaming UI and format db snapshot
Biome was failing CI on migrations/meta/0261_snapshot.json. Also settle
assistant streaming/tool flags when SSE ends without a terminal frame,
without clobbering Stop's finalized content.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(agent-stream): satisfy biome format and import order
Auto-format the sim package for CI lint:check, and repair the Anthropic
streaming tool-loop payload after an unsafe delete-to-undefined rewrite.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(agent-stream): keep drained answer on abort and update migration journal test
Treat AbortError from reader.cancel as a cancelled pump result so soft-complete
retains answerText. Point the workspace storage migration journal assertion at
0261_chat_include_thinking.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(chat): keep Stop notice when server emits cancel error
Ignore terminal SSE error frames after the user aborts so
"Client cancelled request" cannot overwrite "Response stopped by user".
Co-authored-by: Cursor <cursoragent@cursor.com>
* improvement(chat): ChatGPT-style thinking shimmer and stick-to-bottom scroll
Add left-to-right shimmer on live thinking label/body, keep scroll working by
shimmering an inner node, and follow the answer only while near the bottom.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(agent-stream): stop pump on client disconnect; soft-complete agents only
Abort the agent stream pump when the projected HTTP body is cancelled so
provider work does not continue after disconnect. Limit AbortError soft-success
to Agent blocks so Function/HTTP cancels still fail in logs.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(agent-stream): persist includeThinking across pause snapshots
Paused chat runs with Include thinking enabled were dropping the flag when
serializing the pause snapshot, so resume always rebuilt streams without
thinking/tool SSE frames.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(agent-stream): keep drained answer text when stream times out
Persist pump answerText onto the streaming execution before throwing on
timeout, and carry that partial content into the failed block output so
logs match what the client already saw.
Co-authored-by: Cursor <cursoragent@cursor.com>
* improvement(chat): auto-collapse tools chrome when tool streaming ends
Match thinking UX: open while tools run, collapse when finished, and keep
the panel open only if the user manually reopens it.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(agent-stream): settle canvas stream chrome on failure paths
Clear agentStreamActive and settle running tool chips when blocks error,
timeouts cancel runs, or execution ends without stream:done so the output
panel does not stay on live Thinking/Using tools chrome.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(lint): organize imports in terminal console store
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(agent-stream): mark open tools cancelled on HITL pause
Pause can interrupt a tool loop without tool end events; settling those
chips as success incorrectly showed unfinished tools as complete.
Co-authored-by: Cursor <cursoragent@cursor.com>
* chore(db): drop branch-local 0261 migration ahead of staging merge
* chore(db): regenerate include_thinking migration as 0266 post staging merge
* fix(providers): resolve type errors in streaming tool loop call sites
* fix(agent-stream): gate agent events opt-in and correct provider loop behavior
- streamToolCalls and provider thinking requests now require run-level
agentEvents opt-in (canvas on, chat dual-gated, API off) so existing
runs keep pre-agent-events behavior exactly
- OpenAI reasoning summaries opt-in + strip-and-retry on unverified-org 400
- streaming loops run tool postProcess again (firecrawl/exa async results)
- bedrock live loop falls back to silent path for responseFormat
- deepseek: reasoning_content pass-back unconditional, 'none' sends disabled
- groq: x_groq.usage fallback, reasoning params gated, qwen none disables
- gemini: functionCall parts echoed verbatim, local ids only for events
- truncated turns (max_tokens/length) no longer execute partial tool calls
- MAX_TOOL_ITERATIONS exit flushes last turn text as final answer
- iterations reports actual model calls; shared loop plumbing extracted
* refactor(agent-stream): consolidate protocol, dedupe client/server plumbing, hygiene
- canonical ChatStreamFrame union + type guards consumed by server emitters
and the chat client; stream_error restored to legacy log-only handling
- strip thinking/tool args from providerTiming on public final envelopes
- shared tool-chip lifecycle module for chat, canvas, and console store
- shared sink-to-execution-events forwarder replaces the copy-pasted
adapter in the execute route and HITL manager; LIVE_ONLY event set shared
- stream:thinking payload field renamed data->text; canvas thinking batched
- abort reasons carried as AbortError DOMExceptions so raw fetch consumers
classify correctly; thinking cap renamed to chars and scope-documented
- kimi wired for agent events like the other compat providers
- deleted dead exports/step-N comments; fixtures match real wire shapes;
loop tests use explicit mocks instead of importOriginal
* test(agent-stream): cover the dual-gated execution path and typed abort reasons
- chat route tests assert agentEvents reaches executeWorkflow only when
policy and protocol header agree
- execution-limits tests assert AbortError-typed reasons
- executor metadata type carries agentEvents
* fix(deploy-modal): align include-thinking spacing with the modal's 6.5px rhythm
* docs(agent-stream): autogenerate per-model thinking/tool stream support on the Agent block page
- capabilities.thinking.streamed ('full' | 'summary' | 'none') on models.ts,
explicit for the Anthropic family where visibility varies per generation;
getThinkingStreamVisibility exposes the derivation for docs and UI alike
- scripts/sync-agent-stream-docs.ts regenerates the support tables between
markers in workflows/blocks/agent.mdx from the model registry and
STREAMING_TOOL_CALL_PROVIDERS; --check fails on drift or missing metadata
- wired agent-stream-docs:check into CI next to the other sync gates
* feat(anthropic): request summarized thinking display for omitted-default Claude models
The newest Claude generations (Fable 5, Sonnet 5, Opus 4.8/4.7) default
thinking.display to omitted — empty thinking blocks, no deltas. On
agent-events runs Sim now opts back in with display: 'summarized', driven
by the registry's streamed metadata; legacy runs keep the exact
pre-agent-events request shape. Registry, generated docs, and the family
capability table updated accordingly.
* docs(skills): cover thinking.streamed and agent-stream docs sync in model skills
* chore(deps): upgrade @anthropic-ai/sdk to 0.114.0 and adopt official types
- adaptive thinking, display, and output_config are now SDK-typed; the only
remaining custom payload field is output_format (beta-header structured
outputs, which the SDK models as output_config.format instead)
- anthropic stream events narrow on the SDK's discriminated unions instead
of anonymous casts; compat deltas type content/tool_calls from the OpenAI
SDK with vendor reasoning fields as an explicit optional extension
- @sim/auth exposes an explicit VerifyAuth contract so its declarations no
longer reference better-auth's nested zod instance (TS2883 under fresh
install layouts); realtime consumer aligned
- docs app zod pinned to the repo's exact 4.3.6 so ai SDK types bind the
same zod instance (docs type-check was latently broken)
- knowledge embedding tests made hermetic against local .env keys and
hosted rotation fallback
* refactor(providers): replace legacy as-any stream casts with annotated typed casts
* refactor(providers): finish provider audit — remove dead byte-stream helper, annotate remaining legacy casts
Audit of all 26 providers for the agent-events feature confirmed every
streaming execution declares agent-events-v1 and every adapter emits
AgentStreamEvent objects. Cleanup from the audit: the unconsumed legacy
createOpenAICompatibleStream byte helper is deleted, and the remaining
streamResponse-as-any casts (xai, nvidia, kimi, meta, zai, sakana) are
annotated typed casts matching the groq/deepseek fix.
* feat(streaming): stream answer text live during tool loops via turn_end protocol
The live tool loops buffered all answer text per model turn (classification
of intermediate vs final is only known at turn end), so gated surfaces saw
thinking stream, then dead air with the thinking chrome stuck open, then the
whole answer at once.
Loops now emit text deltas live as `turn: 'pending'` plus a `turn_end`
event per turn. The pump buffers pending text and projects it to the byte
path (answerText/logs/memory/legacy clients) only on a final turn_end, so
all settled semantics are unchanged. Gated surfaces render the pending text
as it streams and reconcile with a reset when a turn resolves to tools:
- public chat: live `chunk` frames from the sink + dual-gated `chunk_reset`;
byte-path frame emission is suppressed to avoid duplicates (kept for
response-format transformed streams via clientStreamTransformed)
- canvas: forwarder emits live `stream:chunk` + `stream:chunk_reset`; the
execute route and HITL resume readers stop re-emitting byte chunks; panel
chat tracks per-block segments and replaces content on flush
- chat client: per-block text segments, chunk_reset handling, and thinking
chrome now settles on tool start as well as first answer chunk
* fix(streaming): address validated review findings across provider gating and reset reconciliation
Three-reviewer pass over the branch, findings validated against staging:
- agent-handler forwards agentEvents to executeProviderRequest — the flag was
computed but dropped in the field-by-field copy, so provider-side thinking
requests (OpenAI summaries, Gemini includeThoughts, Anthropic summarized
display) never activated on opted-in runs
- openai: restore summary:'auto' alongside explicit reasoning effort — staging
always paired them; gating summary purely on agentEvents changed legacy
payloads
- gemini: Gemini 2 + tools + responseFormat falls back to the silent path;
the live loop never applied the deferred responseSchema for AUTO tools
- openai-compat loop: malformed tool-argument JSON fails the call instead of
executing with defaulted {} args (staging parsed inside the execution try)
- openai-compat parser: a vendor id arriving after a synthesized start no
longer renames the call (start/end ids stayed consistent)
- stream-pump: abort closes the byte projection so a drain blocked on
backpressure cannot deadlock teardown
- chunk_reset removes the block from the client text order (deployed chat +
panel chat) so a reset block re-registers at arrival position — fixes
separator/order corruption when parallel blocks stream around a reset
- resume route echoes the negotiated X-Sim-Stream-Protocol response header
(parity with the chat route); docs: [DONE] wire shape + final-vs-error
terminal semantics corrected
* chore(deps): exempt pinned @anthropic-ai/sdk 0.114.0 from the release-age gate
CI's bun install --frozen-lockfile blocks 0.114.0 (published 2026-07-23,
younger than the 7-day supply-chain gate). The pin is exact and was vetted
for the agent-events streaming work; following the existing bunfig pattern,
the exclusion ages out on 2026-07-30 and should be dropped then.
* chore(providers): fix double-cast-allowed annotation placement for the strict boundary audit
The audit only recognizes the annotation on the line directly above the cast;
two annotations had drifted behind intervening code lines (groq stream params,
deepseek loop messages) and the OpenAI reasoning-summary widening cast was
never annotated. No behavior change.
* fix(chat): settle straggler tool chips as error when final reports failure
A failed run can still terminate with a `final` frame carrying success: false;
running chips previously settled green regardless of the outcome.
* fix(canvas): wire agent stream chrome into run-from-block
Run-from-block executions emit the same live stream:thinking/stream:tool
events as full runs but registered none of the handlers, so the terminal
never showed thinking or tool chips on that path. The per-run chrome
(batched thinking writes + tool chip lifecycle + settlement on stream done,
block error, and every terminal execution state) is extracted into a shared
createAgentStreamChrome factory consumed by both paths.
---------
Co-authored-by: Bill Leoutsakos <billleoutsakos@Bills-MacBook-Pro.local>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Vikhyath Mondreti <vikhyath@simstudio.ai>
* feat(skills): permissions layer * chore(db): drop skill_member migration 0261 for regeneration on latest staging Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * feat(db): regenerate skill_member migration as 0262 on latest staging Same DDL as the dropped 0261 (skill_member table, enums, indexes, skill.workspace_shared) plus the hand-written write-user backfill, renumbered after staging's 0261. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * feat(db): regenerate skill_member migration as 0263 after staging merge Staging claimed 0262 (strong_storm); same DDL plus the hand-written write-user backfill, renumbered on the merged snapshot chain. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * chore(db): drop skill_member migration 0263 for regeneration on latest staging Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * feat(db): regenerate skill_member migration as 0264 after staging merge Staging claimed 0263 (workflow_fork_sync_excluded); same DDL plus the hand-written write-user backfill, renumbered on the merged snapshot chain. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * make editing skills full page * fix disclaimer * edit access msg * fix lint * chore(db): drop skill_member migration 0264 for regeneration on latest staging Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * feat(db): regenerate skill_member migration as 0265 after staging merge Staging claimed 0264 (fat_ikaris); same DDL plus the hand-written write-user backfill, renumbered on the merged snapshot chain. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix tests * simplify system * fix * fix lint * add mship skills docs * chore(db): drop skill_member migration 0265 for regeneration on latest staging Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * feat(db): regenerate skill_member migration as 0266 after staging merge Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(deps): override zod to 4.3.6 to dedupe nested copies breaking type-check better-auth 1.6.23 and fumadocs-mdx resolve ^4.3.6 to a nested zod 4.4.3, which makes @sim/auth's inferred betterAuth types non-portable (TS2883) and split docs onto a second zod instance. Both ranges accept the repo-wide pinned 4.3.6, so a single hoisted copy satisfies everything. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix lint * feat(skills,tools): fullscreen skill create + shared custom tool editor Moves the rich-markdown and custom-tool editing surfaces out of modals and onto full-page surfaces, and collapses the duplicated chrome behind shared components. Skills - Add /skills/new, a full-page create surface mirroring the skill detail page (CredentialDetailLayout + DetailSection + unsaved-changes guard). "Add to Sim" navigates there instead of opening a modal. - Import moves to a header action (SkillImportButton) backed by a shared readSkillFile helper; the GitHub-URL import and its /api/skills/import route are removed. - Skill name validation is now one shared validateSkillName, replacing three copies of the kebab-case rule and its messages. - The skill editor roster renders through the shared MemberRow instead of re-deriving its identity block, with a locked role control and a lock-reason tooltip explaining inherited workspace-admin access. Custom tools - Extract the canvas modal's schema/code editors into a shared custom-tool-editor module (fields, wand generation, schema helpers), cutting custom-tool-modal.tsx by ~900 lines. - Settings > Custom tools gains a full-page detail sub-view (SettingsPanel + SettingsSection + saveDiscardActions), deep-linkable via ?custom-tool-id. Rows are clickable; delete now lives only in the detail view. - Replace legacy Button/Input/Badge/Label with the chip family, move chip-field chrome into CodeEditor behind an error prop, and delete its dead wand button. Rich markdown field - maxHeight is now opt-in: omit it on a page and the editor grows with its content so the page owns the only scrollbar. Modals pass explicit caps. - The field variant drops to font-weight 400 to match adjacent chip fields. * fix(skills): address review round on create navigation, 409 copy, and editor audit - Skill create navigated using the first element of the upsert response, but that endpoint returns the caller's whole skill list (built-ins prepended) — match the new skill by its workspace-unique name instead. - The suggested-skill 409 toast claimed the skill existed but was not shared and told the user to ask a skill admin. Every workspace member can already see and use every skill, so a 409 only means the name is taken. - Adding an editor emitted the skill_shared event and SKILL_MEMBER_ADDED audit even when onConflictDoNothing skipped the insert on a concurrent add. Gate both on the insert actually returning a row. * chore: format skills-resolver test import * fix(skills,tools): audit fixes — autocomplete boundary, resize clipping, error routing Two real regressions introduced while simplifying the extracted editor: - The schema-param autocomplete's trigger was rewritten to match a trailing identifier, but the completion still split on separators. The two disagreed, so typing `data.ci` opened the menu and selecting replaced `data.ci` whole — eating the member-access prefix. Both now share one SCHEMA_PARAM_WORD regex. - The uncapped markdown field measured its height only on value change while always setting overflow-hidden, so any width change that re-wrapped lines clipped the tail with no scrollbar to reach it. Now re-measures via ResizeObserver. Also from the audit: - Generation writes bypass the code field's change handler, so an open autocomplete stayed over a disabled streaming editor; close it on busy. - Delete failures rendered in the Schema section's error slot on both custom tool surfaces; route them to a toast instead. - Skill create navigated away while still dirty, stranding the unsaved-changes guard's history sentinel so Back landed on an empty create form. - The Description field on skill create never received its error border. - Drop a double-applied opacity-50 (the editor already dims when disabled), a dead try/catch around a non-throwing call that also shadowed the error prop, and a stale reference to a /tools page that does not exist. - Docs still described the removed GitHub-URL import and the old Add Skill dialog; rewrite for the create page and file/paste import. * feat(tools): read-only tool detail, create lands on the new tool, drop dead wand prompt API - Viewers without edit rights could not open a custom tool at all, while the equivalent skill and custom-block surfaces both offer a read-only view. The detail page now takes `readOnly`: editors inert, no Save/Discard/Delete, no Generate. Creating still requires edit rights. - Creating a tool bounced back to the list while creating a skill lands on the new skill. Tools now do the same. The upsert returns the workspace's whole tool list (newest first) rather than just the new row, so the id is matched by title instead of by index — the same trap that produced the skill-create navigation bug. - Remove `openPrompt`/`closePrompt` from useWand. `closePrompt`'s last callers went away with the custom-tool-modal extraction and `openPrompt` had none before it; nothing reads `isPromptVisible` any more either. * fix(tools): read-only editors, design-system wrench, skills-matching tool identity - readOnly never reached the editors: the prop gated actions and Generate but the schema and code fields were still typable for viewers without edit rights. Wire disabled through both fields into CodeEditor. - The row icon used lucide's Wrench (strokeWidth 2) where @sim/emcn/icons ships one drawn for this system (1.55, tuned viewBox), and it inherited body text colour instead of --text-icon. Swap it. - Give the tool detail page the same identity heading as skill detail: tile, name, and description at the top left, instead of only a header title. - Extract ResourceTile so the skills and tools tiles share one definition (SkillTile now composes it), and add an opt-in `iconFilled` to SettingsResourceRow so the tools list tile matches the skills gallery. Both default to today's behaviour for every existing consumer. * fix(mentions): use the product's own glyph for every @ mention kind The `@` menu and the inserted chip mapped kinds to arbitrary lucide icons — `Sparkles` for a skill, a generic `File` for every file — while the rest of the product has a settled glyph per resource. Mirror CHAT_CONTEXT_KIND_REGISTRY, which Chat's `@` menu already renders from: - skill now uses AgentSkillsIcon, the same glyph SkillTile shows everywhere - workflow / folder / table / knowledge use the @sim/emcn/icons set the sidebar and the chat registry use - file derives its icon from the filename extension, so a .pdf and a .csv are distinguishable, matching the file list and Chat's context chips - integration keeps the block's brand icon from the registry Also drop the generic placeholder. `kind` is untrusted — the node schema defaults it to `''` and a hand-written `sim:` link can carry anything — but an unrecognized kind now yields no icon instead of a meaningless box, which is what the chat registry does. The menu already guarded a missing icon; the chip now does too, so this cannot crash on a malformed link. * chore(db): drop skill_member migration 0266 for regeneration on latest staging * feat(db): regenerate skill_member migration as 0267 after staging merge --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Co-authored-by: Waleed Latif <walif6@gmail.com>
…face (simstudioai#5916) * refactor(skills): one definition of the skill fields across every surface The Name / Description / Content trio was implemented three times — once in the canvas modal and once each in the create and detail pages — and had already drifted: the name hint appeared on create, was missing entirely on detail, and sat in a different slot in the modal; the content editor was 260px on the pages and 200px in the modal; only the modal marked the fields required. Extract SkillFields, the DetailSection form both full-page surfaces render, and move the shared copy into skill-copy.ts. The modal keeps ChipModalField — that is required inside a ChipModalBody, so it cannot share the pages' JSX — but now reads the same placeholder, hint, and max-length constants, so the wording can no longer diverge. The detail page's local FieldLockTooltip moves into SkillFields and is now available to both pages, and the dynamic rich-editor import drops from three copies to two. No behavioral change beyond the drift being resolved: the name hint now shows on the detail page as it already did on create. * refactor(skills): derive modal saving state, collapse the field message line From the cleanup passes over this branch: - SkillModal mirrored `mutation.isPending` into a local `saving` useState, which .claude/rules/sim-hooks.md forbids. It also carried a latent bug: the `finally { setSaving(false) }` ran after `onSave()` had closed the modal, so it set state on an unmounting component, and a throw from `onSave()` would leave the modal stuck in a saving state. Derived from the two mutations instead. - The hint/error line under a field was written three times in SkillFields, in three slightly different shapes. One FieldMessage helper now renders all three. - The name hint is an authoring instruction, so it no longer shows under a field that cannot be edited (a built-in skill, or a viewer who is not an editor).
…y chat answers) (simstudioai#5915) * fix(providers): stop the final regenerated stream from re-calling tools and clobbering the answer After the silent tool loop settles, OpenAI Responses and Gemini re-issue a streaming request purely to stream the answer as prose — but with tools still attached and auto tool choice, a reasoning model can re-decide to call a tool there. Streamed calls are never executed on this path, so the run ends with a dead function call, an empty streamed answer, and the stream callback clobbering the tool loop's settled text with '' (deployed chat rendered {"content": ""}). Force tool_choice 'none' / functionCallingConfig NONE on the regeneration and keep the tool loop's settled answer whenever the stream ends without text. * feat(openai): settled tool chips on the regenerated answer stream The silent Responses tool loop has no live stream while tools run, so opted-in consumers saw no tool chips at all for OpenAI. The loop now records each executed call and prepends settled tool_call_start/end pairs (name + status only) to the agent-events stream ahead of the regenerated answer. Runs without a sink never see these events, so legacy output is unchanged. * fix(providers): apply the regeneration guard fleet-wide Audit of every silent tool loop for the same race fixed for OpenAI/Gemini (final regenerated stream re-calls a tool that is never executed, ending with an empty answer that clobbers the settled text): - anthropic (both implementations): tool_choice {type:'none'} on the regeneration (tools must stay — history carries tool_use blocks) + keep the settled answer when the stream ends without text - groq: was re-applying the ORIGINAL tool_choice, so forced-tool runs re-forced the tool on the regeneration — guaranteed dead call; now 'none' + fallback - deepseek, mistral, cerebras, azure-openai (legacy chat path), openrouter, xai: 'auto' -> 'none' + fallback - bedrock: fallback only — Bedrock's ToolChoice has no 'none' and toolConfig is required when history carries toolUse blocks Already guarded (no change): meta, sakana, nvidia, vllm, litellm, baseten, together, fireworks, kimi, zai, ollama.
Co-authored-by: Sim Pi Agent <pi@sim.ai>
…wedged saves (simstudioai#5905) * fix(data-retention): clamp sub-day retention values so saves aren't wedged A stored value under 12 hours rounded to '0' days on load and was re-sent as 0, which the contract rejects (min 24) — blocking every save on the page, including unrelated fields. Clamp hours->days into the contract's range on read, and throw instead of emitting 0/NaN on write. * improvement(docs): rewrite data retention for workspace overrides + PII redaction - Document PII redaction (Logs / Workflow input / Block outputs stages, entity types, languages, custom regex patterns) - Replace the stale 'no per-workspace overrides' section with the retention-policies list and override inheritance - Correct log retention (also covers background job logs) and soft deletion (adds Chat conversations, KB documents) - Add PII + override screenshots, refresh the main one
…udioai#5909) * feat(sso): DNS domain verification gating org SSO registration Add org-scoped domain ownership verification (DNS TXT challenge) as the security precondition for configuring SSO. Closes the first-come domain-claim vuln where any org could wire another company's domain to its own IdP. - New sso_domain table + migration 0266; existing org SSO domains are grandfathered as verified so live tenants are unaffected - Verified-domains settings UI (enterprise-gated) with add/verify/remove - Register route now requires a verified domain for org-scoped registration; personal SSO and already-grandfathered domains are unaffected - Self-host register script writes the verified sso_domain row directly, so script-driven registration stays backwards compatible * fix(sso): harden domain verification against concurrency + fix CI lint Addresses review findings on state invariants under concurrent/failed writes: - Add unique index on (organization_id, domain) so concurrent claims can't create duplicate pending rows; POST re-reads and stays idempotent on conflict - Verify flips the row only if it's still the exact pending challenge checked (guards deletion/token-rotation mid-DNS-lookup) and maps the partial unique index violation to 409 instead of an unhandled 500 - Wrap the self-host script's provider write + verified-domain upsert in a transaction so a failed ownership write can't leave a provider committed - Format 0266 snapshot/journal with biome (fixes @sim/db lint:check) * fix(sso): re-check domain verification before provider write (TOCTOU) The register gate checked the verified sso_domain row only at handler entry, then ran OIDC discovery before writing the provider. A verified row removed during that window could still complete registration. Extract the check into a closure and call it both as an entry fast-fail and authoritatively right before registerSSOProvider, alongside the existing domain-conflict re-check. * fix(sso): stop rotating verification token on idempotent re-add Re-adding a pending domain rotated its verification token, which invalidated a TXT record the admin may have already published and — under two concurrent re-adds — could return a token the racing write had already superseded, so the admin's DNS record would never verify. Return the existing row unchanged instead; the pending token is always shown in the UI, so it is never lost. * fix(sso): close register TOCTOU with compensating delete + harden edges Audit-driven hardening: - Close the residual register TOCTOU: registerSSOProvider is create-only (throws if the providerId exists), so a compensating delete after the write is provably safe — it can only remove the just-created row. If verification was revoked during the write, roll the provider back and 403. - Verify is now idempotent under concurrency: a same-org row already flipped to verified by a racing request returns 200, not a confusing 409. - Grandfather backfill + self-host script now match normalizeSSODomain's dominant transforms (lower + trim + strip leading wildcard) so a non-canonical legacy domain can't miss the runtime gate's lookup. Prod backfill result is unchanged. - Cleanup: drop dead default export, align card radius to sibling convention. * fix(sso): redact domain tokens from non-admins + fix script stale-update Round-5 review findings: - GET /domains redacted the pending TXT verification token (a management secret) to any org member. Now only owner/admins read it; members see the list and status without it. Non-Enterprise orgs get an empty list (entitlement flag only), never the domains/tokens. - Self-host script decided update-vs-insert from a read taken OUTSIDE the transaction; a provider deleted mid-flight made the UPDATE match zero rows silently while the verified-domain upsert still committed (orphaned domain). The decision now happens inside the transaction from the UPDATE's row count. * docs(sso): drop unshipped enforce-SSO / auto-join copy from verified domains Verified domains currently only gate SSO configuration. Remove the forward-looking references to enforcing SSO and auto-joining members (deferred to a later release) from the docs, settings copy, nav description, and schema comment so we don't promise unshipped features. * fix(sso): guard rollback to new providers only + Enterprise-gate domain removal Round-6 review findings: - The compensating provider rollback now only fires when the provider did not exist before this request (providerExistedBefore). registerSSOProvider is create-only today so reaching the rollback already implies a fresh create, but this makes the safety local and future-proof: if Better Auth ever allowed updating an existing provider, a revoked-verification rollback must not delete that pre-existing row. - DELETE /domains now requires an Enterprise plan like add/list/verify, so all domain mutations share one entitlement (the UI already hides removal from non-Enterprise orgs). Adds a delete-route test. * fix(sso): roll back the SSO provider by row id, not logical keys The compensating rollback deleted by (providerId, orgId). providerId is unique, so if this request's row were deleted and recreated by a concurrent registration in the narrow window before the rollback, the logical-key delete would remove that other request's provider. Delete by the primary-key id registerSSOProvider returns instead, so only the exact row this request created is ever removed. * chore(sso): final-review polish — trim script read, unify copy, doc migration edge Cosmetic cleanup from a final 4-track adversarial review (no bugs found in the new logic): - Self-host script: narrow the pre-transaction existence read to select({ id }) instead of SELECT * (it only feeds a log line now). - Unify invalid-domain copy ("for example acme.com") and the verified-elsewhere 409 wording ("is already verified by another organization") across routes. - p-3 shorthand on the domain row card. - Document the migration's rare two-orgs-share-a-domain grandfather behavior (login unaffected; validated no such duplicates in prod). * fix(sso): apply attribute mapping + make SSO edit work; drop dead guard Two pre-existing SSO bugs the final review surfaced (prod has one SSO org, RVW, script-registered with the default mapping, so neither change affects it): - Attribute mapping was passed at the top level of the register payload, which Better Auth ignores — it reads oidcConfig.mapping / samlConfig.mapping. Nest it so custom mappings actually apply. (Default mapping is unchanged, so existing logins are unaffected.) - Editing an SSO provider was broken: registerSSOProvider is create-only and threw on the existing providerId → generic 500. Route now detects a provider the caller already owns and updates it via Better Auth's updateSSOProvider, and surfaces Better Auth's own error status/message instead of a blanket 500. Also drops the now-unnecessary providerExistedBefore guard (the rollback deletes by the created row's primary-key id and register is create-only) and the earlier final-review polish (script read, unified copy, migration edge note). Smoke-test SSO login + edit on staging before merge (auth-path change). * fix(sso): require null org on personal-mode provider lookups (gate bypass) The personal branch of both provider-ownership lookups keyed on (providerId, userId) without requiring organizationId IS NULL. Because org providers store userId = their creator and providerId is globally unique, an org admin could send a personal-mode request (no orgId) — which skips the membership check and the domain-verification gate — yet still match, and then via the new update path move, their org's provider to an unverified domain. Add isNull(organizationId) to the personal branch of both clauses so it can only match a genuinely personal provider, matching the route's own isOwnedByCaller. Found by an adversarial review of the update path added in 394bda9. * fix(sso): script updates the observed provider by id, not providerId Inside the registration transaction the script updated WHERE providerId — the logical key. If the observed provider was deregistered and a replacement created with the same providerId before the transaction ran, that update would clobber the replacement's config and ownership. Update the specific observed row by its primary-key id instead; if it's gone we insert, which fails cleanly on the providerId unique constraint rather than overwriting the replacement. * fix(sso): script upserts provider via delete-then-insert (no unique constraint) sso_provider.provider_id is a plain (non-unique) index and prod holds legitimate duplicates, so the previous "update by id, else insert" could create a duplicate provider when the observed row was deregistered and replaced before the transaction — the fallback insert would succeed. Delete every row for the providerId then insert exactly one, inside the transaction, so the providerId ends up as exactly this config atomically. Linked accounts key on the providerId string (not the row id), so existing logins are unaffected. * fix(sso): guard compensating-delete row id so rollback can't silently no-op * chore(sso): regenerate migration as 0268 after merging staging Staging landed migrations 0266/0267, colliding with our 0266. Removed our migration, merged staging, and regenerated cleanly with drizzle-kit as 0268_sso_domain_verification (identical sso_domain table + indexes), then re-appended the grandfather backfill. api-validation baseline reconciled to 973 (staging 970 + our 3 domain routes). Also make the register-route test's registerSSOProvider mock return an id so the guarded compensating delete runs. * refactor(sso): share normalizeSSODomain via @sim/utils so script matches gate The self-host script canonicalized SSO domains with a minimal inline transform (lower+trim+wildcard) that diverged from the app's full normalizeSSODomain (protocol, port, path, trailing dot, email local part) — equivalent spellings could store a different ownership key than the runtime gate looks up. Move normalizeSSODomain into @sim/utils/sso-domain (a pure function) so the register route, the domain-claim route, and the script all use the identical canonicalizer. The script now skips the verified-domain record when SSO_DOMAIN isn't a valid registrable domain instead of storing a malformed key.
* feat(providers): add Claude Opus 5 model * fix(providers): route Opus 5 through adaptive thinking Opus 5 supports only adaptive thinking (no extended thinking / budget_tokens), same as Opus 4.8/4.7. Add opus-5 to supportsAdaptiveThinking so thinking requests use thinking.type: adaptive instead of falling through to budget_tokens extended thinking, which Opus 4.7+ rejects with a 400. * docs(agent): regenerate agent-stream docs for Opus 5
…ides (simstudioai#5928) * content(library): add observability, procurement, and MCP security guides Three AEO guides, each dated into a recent empty day in the posting calendar (2026-07-19, 07-21, 07-22) so publication is spread across days rather than dumped on one. - ai-agent-observability: what observability is, why APM falls short for non-deterministic agents, what to instrument per lifecycle stage, and the signals to track - ai-agents-in-procurement: what procurement agents do, buy-vs-build, where they add value, and how to start narrow - mcp-security: tool poisoning, confused-deputy/OAuth flaws, and supply-chain risk, plus how to build and govern MCP servers securely Every third-party factual claim carries an outbound citation to a verified primary source (OpenTelemetry semconv, Fiddler, LangChain and PwC surveys, Icertis/ProcureCon, Ironclad, GEP, the MCPoison/CurXecute CVEs on NVD, Anthropic's MCP announcement, RFC 8707/9700, OAuth 2.1, the MCP auth spec), and each post carries 3 internal links to related library posts. One claim from the source copy — a "November 2025" date on the WhatsApp MCP exfil case — could not be verified and was dropped; the case itself is cited to Docker's writeup. Covers come from the autogeneration pipeline via the standard /library/<slug>/cover.jpg path. * fix(library): correct two unverifiable claims in the new guides Accuracy pass on the three new posts turned up two claims that could not be substantiated: - HIPAA compliance: the procurement post claimed Sim has "SOC2 and HIPAA compliance," but the canonical compliance data (lib/compare/data/sim.ts) states SOC2 only, and explicitly that Sim offers self-hosting "beyond SOC2, rather than additional certifications." Removed HIPAA; kept SOC2 plus self-hosting for data residency. (The same claim exists in ~5 pre-existing library posts and should be corrected separately.) - "more than 18,000 servers were listed on MCP Market": no source substantiates this figure. Replaced with "thousands of community-built servers," which the ecosystem supports, keeping the Anthropic and MCP Market links. Every other third-party claim was verified against a primary source (both CVEs on NVD, RFC 8707 title, RFC 9700 as January 2025, and all six survey statistics).
…uite (simstudioai#5927) * improvement(ship): pre-flight regenerate artifacts + parallel audit suite Add a two-phase pre-ship step: (A) regenerate every committed artifact (agent-stream-docs, skills, contract syncs) in parallel so generated files never drift into a CI failure, then (B) run lint plus the full audit suite CI's Lint and Test job enforces, in parallel, aborting ship on any failure. Regenerate the ship command projections. * fix(ship): scope Phase A to in-repo generators + propagate failures Cursor review: (1) mship:generate is an umbrella over the 9 contract generators and biome-formats the shared generated dir, so running it alongside its constituents races/corrupts; it also reads an external copilot-contract source that ENOENTs in a bare worktree. (2) bare wait swallowed generator exit codes. Narrow Phase A to the always-in-repo generators (agent-stream-docs, skills) and collect each job's exit status in both phases; document domain generators as on-demand only. * fix(ship): make pre-flight status checks exit non-zero on failure Cursor/Greptile: grep && echo ❌ || echo ✅ always exits 0, so a failed generator/audit didn't actually gate ship (regressed the sequential bun-run checks whose non-zero exit agents rely on). Replace both Phase A and Phase B aggregations with an if-grep that exits 1 on any failure. * fix(ship): gate lint exit in Phase B pre-flight Cursor: bare bun run lint before the audit grep meant a non-zero lint was ignored — if audits then passed, the block exited 0 and ship continued. Gate lint with || { echo …; exit 1; } so an unfixable lint error aborts before the audits run.
…end gotcha (simstudioai#5931) * fix(sso): surface DNS verification failures and the provider auto-append gotcha Post-merge audit follow-ups for verified domains (simstudioai#5909): - The host field handed admins the FQDN `_sim-challenge.acme.com`. GoDaddy, Namecheap, Hover and most cPanel panels append the zone to whatever is typed, yielding `_sim-challenge.acme.com.acme.com` — the record looks right in their panel but never verifies, and our 422 tells them to wait 48 hours. Add a hint under the field (and a docs callout) telling those admins to enter just the label. - DNS failures were logged at debug, but production log level is ERROR, so a blocked-egress or SERVFAIL condition was invisible to us and misreported to the admin as "record not found yet". Log infrastructure-class failures at warn with the DNS error code; keep the genuinely-absent codes at debug. - Trim the joined TXT value before comparing: several DNS panels pad the stored string, which otherwise fails an exact match forever. - Lower the resolver to 2s/1 try. c-ares multiplies timeout across servers and retries by ~7x, so the previous 5s/2-try config could block a verify request for ~35s when resolvers are unreachable. - Cover `checkDomainTxtRecord` — the function that decides whether the gate opens had no tests. Adds exact match, chunk-joined value, match among unrelated records, padded value, near-miss, another org's token, absent record, infrastructure failure, and empty-response cases. * fix(sso): log DNS infrastructure failures at error so prod actually surfaces them Production's default minimum log level is ERROR, so the warn introduced in the previous commit was still filtered out — the fault stayed invisible exactly as before. A resolver failure that is not 'record absent' is a genuine infrastructure error, so ERROR is both the visible and the honest severity. * fix(sso): state the zone-removal rule instead of a wrong subdomain hint The hint computed the bare label as the first segment of the challenge host, so for a subdomain like eng.acme.com it advised entering `_sim-challenge` when the host relative to the acme.com zone is `_sim-challenge.eng` — following it would publish the record on the wrong name and verification would never succeed, the exact failure the hint exists to prevent. Deriving the real zone needs the Public Suffix List (acme.co.uk defeats naive label-stripping), so state the rule instead: enter the host with the trailing zone removed. Docs show both the apex and the subdomain form.
…imstudioai#5935) .npmrc set ignore-scripts=true globally, which overrode Bun's trustedDependencies allowlist and blocked every lifecycle script — including the repo's own root scripts. isolated-vm therefore shipped unbuilt and each contributor had to npm rebuild it by hand. The .npmrc line landed in the May 2025 Bun migration, seven months before isolated-vm was introduced. Two later commits added isolated-vm to trustedDependencies (root, then apps/sim) and both were silently inert. - delete .npmrc; lifecycle policy now lives solely in root trustedDependencies - add isolated-vm to root trustedDependencies - drop the workspace-level trustedDependencies from apps/sim (Bun only reads the root list, so it never had any effect) - drop the .npmrc entry from CODEOWNERS Naming any package in trustedDependencies replaces Bun's curated default-trust list, so only isolated-vm and sharp may run install scripts. Docker is unaffected — all three Dockerfiles pass --ignore-scripts explicitly and rebuild isolated-vm against the Node ABI by hand.
…mstudioai#5936) Reconcile compliance and traction wording across three library guides against the canonical data in lib/compare/data/sim.ts, and tighten a couple of related phrasings for consistency.
…O v1 default (simstudioai#5939) * improvement(helm): hygiene pass — CI gating, strict values schema, ESO v1 default, ci values Closes the gaps from a best-practices audit of the chart (template-level conformance was already clean: full label set, 82 unit tests, kubeconform- valid renders): - new Helm Chart workflow gates every chart change: helm lint, the 82 helm-unittest cases, kubeconform validation of default + all-components renders (k8s 1.29 strict, CRD catalog), and a render of all 10 example values files - helm/sim/ci/ values files (chart-testing convention) so the chart lints and templates cleanly out of the box with dummy secrets - values.schema.json declares all 30 top-level keys (16 were invisible) and sets root additionalProperties: false, so top-level typos fail fast - externalSecrets.apiVersion defaults to v1: current ESO releases removed the v1beta1 compatibility path in 2026, so the old default produced rejected manifests on new installs; NOTES/values comments updated - wait-for-postgres init container gets requests/limits (the only container in the chart without them; broke ResourceQuota'd namespaces) - drop the telemetry Prometheus scrape config for app/realtime — neither exposes /metrics, so it was dead config that also rendered a realtime target with realtime disabled - chart 1.2.0 with README upgrade notes * fix(helm): version-agnostic schema-error assertions in the secret-length suite Helm v4 phrases schema rejections as 'minLength: got N, want 32' while v3 says 'String length must be greater than or equal to 32'. The suite grepped for 'minLength' only, so it passed on local helm v4 and failed on CI's v3.16.4 — the enforcement itself works on both. Patterns now assert the key name plus either wording. Caught by the new Helm Chart workflow on its very first run. * improvement(helm): kind install test + chart version-bump gate in CI Benchmarked against the flagship OSS charts (ingress-nginx, argo-cd, kube-prometheus-stack, grafana, bitnami, cert-manager): sim already exceeds most of them on validation rigor (strict schema — 4 of 6 ship none; 82 unit tests vs argo-cd's zero; kubeconform manifest validation none of them run), but every top community chart repo actually installs the chart on a kind cluster in chart CI — the one majority practice we lacked. Adds: - install job: kind cluster, helm install with ci/default + a new small-footprint ci/kind-values.yaml overlay (default app requests of 4Gi can't schedule on a CI node), --wait, then the chart's helm test hook, with pod/event/log diagnostics on failure - version-bump job (PR-only): fails when helm/sim/** changes without a Chart.yaml version increment — argo/kps/grafana all enforce this * fix(helm): declare naming overrides in the strict schema; SHA-pin CI actions - nameOverride/fullnameOverride are consumed by sim.name/sim.fullname but were never in values.yaml, so root additionalProperties: false rejected Helm's standard naming overrides — both now declared (a helper-wide sweep confirmed they were the only template-read keys missing), with a smoke regression test that installs under the strict schema and asserts the override lands in resource names - CI supply-chain hardening: checkout/setup-helm/kind-action pinned to full commit SHAs (tag comments retained) and the kubeconform archive verified against its published sha256 * fix(helm): immutable unittest runner and fail-closed version gate in CI - helm-unittest runs via the project's official docker image pinned by immutable sha256 digest instead of a plugin install from a mutable git tag - the version-bump gate fetches the full base ref (a --depth=1 fetch could leave no merge base), computes the merge-base and diff outside the if so any git failure fails the job instead of falling into the skip branch * fix(ci): run the helm-unittest container as the runner UID The digest-pinned image runs as a non-root user that cannot write into the runner-owned bind mount (it creates tests/__snapshot__, absent from the checkout since empty dirs aren't tracked). Standard bind-mount pattern: --user "$(id -u):$(id -g)" with HOME=/tmp for helm's cache. * fix(helm): appVersion points at a real GHCR tag (v0.7.44) The kind install job caught this on its first full run: the default image tag (Chart.AppVersion 0.6.73) returns 404 on GHCR for all three images — the registry's tags are v-prefixed — so an unpinned default install could never pull. Updated appVersion to v0.7.44 (verified 200 for simstudio, realtime, and migrations manifests), stale values comment refreshed, and an upgrade note added. The CI kind values deliberately stay tag-free so the job keeps exercising the true default path.
simstudioai#5922) * improvement(providers): validation pass, and stream tool loop improvements * remove deploy options correctly * fix * feat(providers): prompt caching capability and usage-based cache pricing Replace the arbitrary cached-rate heuristic with a single cache-aware pricing function, and add prompt caching as an opt-in capability for Anthropic. Pricing: priceModelUsage in cost-policy.ts is now the only place cache arithmetic happens. Provider adapters normalize their wire shape into ModelUsage (input always excludes cache buckets); the pricing function never branches on provider. This removes five divergent behaviors, including the !!request.context heuristic that gave Router and Evaluator an unearned 10x input discount, and the overwrite that silently billed Anthropic cache reads and writes at zero. Also parses OpenAI cache_write_tokens, previously ignored. Caching: Anthropic gets a capability-gated advanced switch that places cache_control on the last tool and last system block; system is now always a TextBlockParam array. OpenAI gets a stable per-block prompt_cache_key with no UI, since its caching is automatic. * fix(providers): route OpenAI and Gemini block cost through cache-aware pricing Cache-aware pricing only reached trace segments. The billable block cost still called calculateCost on the cache-inclusive prompt total, so OpenAI cache hits and Gemini implicit-cache hits were charged at the full input rate and GPT-5.6+ cache writes went unbilled. Both providers now accumulate cache buckets and price through priceModelUsage, matching the Anthropic token convention where input excludes cache reads and writes. Cached counts are clamped to the prompt total so an over-reporting payload cannot bill more input than the request contained. * fix(streaming): redact tool payloads on selected outputs in public chat Redaction only ran on the empty-selection branch, but a deployment almost always selects outputs, so it was dead in the case it exists for. Selecting toolCalls streamed the raw arguments and results to a public chat client in a chunk frame, and providerTiming carried thinking content the same way. Both paths now extract from the sanitized block output rather than the raw log: the streamed selected output, which is the reachable vector, and the final envelope. Sanitizing the source rather than per selected path means a newly selectable field cannot reopen the hole. * refactor(providers): drop unreachable billing fallbacks Every provider pricing helper took a policy parameter no caller passed. Worse than dead: passing one would have double-applied the margin the central layer already applies. Removed, so providers can only price at list. Also removed guards that cannot fire. The central fallback normalized cache buckets no provider can reach it with (all three that report cache usage price themselves) and did so at a 1x write multiplier no vendor charges. priceModelUsage re-validated token counts the adapter had already clamped, and applyModelCostPolicy defaulted a required total field. Validation now happens once, in the adapter that parses the vendor payload and is the only layer that knows cache buckets are a subset of the prompt total.
…nner sizing (simstudioai#5945) * fix(ci): key node_modules sticky disk on the lockfile hash * improvement(ci): per-image Blacksmith runner sizing + cold-build memory preflight * fix(deps): exclude @next/swc binaries from the release-age gate * fix(deps): pin @next/swc binaries so frozen installs get a compiler * docs(ci): explain ARM runner sizing rationale
…inputs/outputs (simstudioai#5942) * improvement(whatsapp): validate + improve integration skill for file inputs/outputs * fix lint * add whatsapp subblock migration
simstudioai#5671 rewrote the shimmer-sweep keyframes to run 100% -> -100%, which is already left-to-right, but left the `reverse` from simstudioai#5650 on the animation shorthand. The two flips cancel into a right-to-left sweep on every ShimmerText consumer: subagent labels, tool-call rows, and the new agent-stream thinking chrome. Drop `reverse` so direction lives in exactly one place — the keyframes.
…imstudioai#5917) * fix(realtime): evict revoked collaborators from live workflow rooms via periodic read-access re-validation * fix * fix(realtime): close join/eviction race and make sweep cleanups independent Re-authorize immediately before socket.join so an in-flight join cannot reverse a sweep eviction, and run the sweep's best-effort cleanups independently so a room-state failure cannot skip the presence broadcast. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(realtime): single-flight role resolution and retry failed eviction cleanup Coalesce concurrent role resolutions per (user, workflow) so a slow stale read can never overwrite a recorded revocation, and defer failed eviction room-state cleanups into a per-sweep retry queue so collaborators are not left with a stale presence entry. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(realtime): scope room removal to the target workflow and detect swallowed cleanup failures Honor the workflowIdHint as the target room in both room managers so removing a stale room cannot clobber the mapping of a room the socket has since moved to, and confirm eviction cleanup via the returned workflowId plus an unswallowed mapping read so Redis failures actually defer into the retry queue. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(realtime): treat unconfirmed removals as failures and refresh role cache on fresh verify Treat any null removal result as a failed cleanup (the sweep always passes the target room, so null only means failure — including with expired mapping keys), move the same-room rejoin guard to a synchronous check immediately before the removal, and record verifyWorkflowAccess's fresh decision into the role cache so a re-granted user is not blocked by a stale cached revocation. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(realtime): prefer a mid-flight recorded role decision over the in-flight query result If a fresh authoritative read (join-time verify) records a decision while a single-flighted resolution's query is in flight, keep the recorded decision instead of overwriting it with the potentially stale result. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(realtime): isolate the security scan from the Redis cleanup lane Run the revocation scan (local sockets + DB only) and the best-effort room-state cleanup as independently-guarded lanes so a hanging Redis command can stall only presence cleanup, never revocation enforcement. Evictions now enqueue cleanup instead of awaiting it, and the scan no longer reads presence for a fallback role. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(realtime): bound authorization waits in the revocation scan Race each socket's authorization check against a per-socket timeout and cap the whole pass with a budget below the sweep interval, so a hanging DB query skips that socket for the pass (never evicting on uncertainty) instead of wedging the scan lane and starving subsequent ticks. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(realtime): round-robin the revocation scan so hung checks cannot starve later sockets Resume each scan pass after the last target the previous pass processed, so a fixed prefix of hanging authorization checks can never repeatedly consume the pass budget and leave sockets behind it unexamined. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
… agent thinking/tool-calls/prompt-caching, opus 5, daytona failover
… state to nuqs, design-system cleanup (simstudioai#5950) * refactor(settings): fold verified domains into SSO and use shared primitives Verified domains only gates SSO, so managing it on a separate page meant discovering the requirement after filling out the whole IdP form and then navigating away mid-setup. Move it into the SSO page as a section above the provider config and drop the standalone page, its nav entry, and both route branches. Align the surfaces with the shared settings primitives rather than bespoke chrome, matching whitelabeling/custom-blocks/access-control: - SSO's local FormField (muted labels) is replaced by the shared SettingRow, so its fields read like every other settings page. SettingRow gains optional `optional` and `error` props to absorb what FormField did — additive, so existing consumers are untouched. - The domains section is built from SettingsSection, SettingRow, SettingsResourceRow, and SettingsEmptyState instead of hand-rolled cards. Also drop the redundant Upload/Change buttons in whitelabeling: the logo and wordmark thumbnails were already clickable, so the button was a second control for the same action. Remove still appears once an image is set. * fix(settings): move group-detail view state to nuqs and clean up design-system drift Access control's group detail kept its tab, three search boxes, and three status filters in useState, so a `?group-id=` link always landed on General and a filter was lost on reload — the parent already puts the group id in the URL. The three tabs never render together, so search and status share one param each rather than carrying three mutually-exclusive keys, and switching tabs resets both. Closing the detail clears all three alongside group-id in one batched write, so nothing lingers on the list URL. Design-system fixes from a cleanup pass over the surfaces this branch touched: - Restore accessible names lost when the whitelabeling Upload buttons were removed. The thumbnail is now the only click target, and it contained just an icon, so it announced as an unlabeled button; the icon-only Remove had the same problem. Both now carry aria-labels reflecting their state. - Use Chip, not the legacy Button, for the domain actions — Button is ~26px against the 30px ChipInput beside it, so "Add domain" sat visibly short. - SettingRow now uses the emcn Info component; a bare svg as Tooltip.Trigger was neither focusable nor nameable. It also stops re-specifying Label's own default styling. - Use the new SettingRow error prop for the group name instead of a hand-rolled error paragraph, which is what the prop was added for. - Hoist the block-category lookup out of a sort comparator, size-* over h/w, name the staleTime constants the rules require, and import RowActionsMenu from its barrel. * fix(settings): alias the old /settings/domains path to SSO Folding verified domains into the SSO page dropped /settings/domains, so bookmarks and shared links 404'd instead of landing where domains now live. Both alias maps already exist for exactly this (organization/'members', subscription/'billing'); add domains -> sso to each. * chore(settings): adopt ChipCopyInput, named staleTime constants, and a11y labels * fix(settings): reset group detail params on open and drop issuer mono styling
…hain (simstudioai#6235) Bumps next, @next/env and the @next/swc-* optional deps to 16.3.0 across the root overrides, apps/sim, apps/docs and packages/emcn. Two config notes worth keeping: - `experimental.turbopackFileSystemCacheForBuild`'s default flipped false -> true for stable in 16.3.0, so our explicit `false` is now load-bearing rather than defensive. Without it this bump would have silently re-enabled a build cache measured 3.2x slower on this codebase (simstudioai#6078). Comment updated to say so. - `experimental.useTypeScriptCli: true` is now pinned. TypeScript 7 ships no JavaScript compiler API until 7.1, so Next's default checker cannot run and needs the project-local `tsc` CLI instead. 16.2.12 was silently skipping build-time type checking entirely because it detected @typescript/native-preview and short-circuited the stage ("Finished TypeScript in 138ms"); pinning the flag keeps that from drifting back. TypeScript toolchain cleanup that the upgrade makes possible: - Drop @typescript/native-preview from apps/sim and apps/docs. It was the pre-release channel for TS 7 and is superseded by typescript@7 (nightlies now ship as typescript@next), and its presence is what suppressed build type checks. - Align packages/browser-protocol and packages/terminal-protocol from typescript ^5.7.3 to ^7.0.2 so every workspace is on one compiler version. apps/sim keeps @typescript/typescript6 as a production dependency: the function sandbox at app/api/function/execute dynamically imports the TS 6 compiler API to transpile user code, and TS 7 has no API to replace it yet. Build and dev were benchmarked 3x per version on a byte-identical tree; the upgrade is performance-neutral (build median 100s -> 99s, dev:full warm 10s -> 9s).
…ath (simstudioai#6232) * fix(providers): route 6-10MB attachments to the provider large-file path The inline base64 cap was 10 MB of raw bytes, but the execution payload store refuses a single value above 8 MiB and base64 inflates by 4/3. Every raw file over 6 MiB therefore produced a base64 string the store rejected — and the rejection came from the base64 *cache* write, which threw and failed the run with "Execution memory limit exceeded" even though the bytes had already been read successfully. Because shouldUseLargeFilePath only fires above the inline cap, 6-10 MB attachments had no path at all on any provider: they never reached the OpenAI or Gemini Files API upload they were supposed to take. Derive the cap from the payload-store ceiling instead of hardcoding it, and degrade a refused cache write to "not cached" rather than failing a request whose bytes are in hand. Every other size guard in the chain compares raw bytes against maxBytes; only the Redis write sees the encoded size, which is why this went unnoticed — and why it failed only where Redis is configured. Also correct the provider ceilings against the vendors' current documentation: - openai: 50 MiB -> 50,000,000. The gate is `size > maxBytes`, so 50 MiB admitted 52,428,800 bytes; the docs say each file must be *under* 50 MB. - bedrock: had no entry and inherited the inline cap, which is above what Converse accepts (3.75 MB per image, 4.5 MB per document). - groq: 20 MiB -> 20,000,000, and modelled as the request cap the docs actually describe rather than a per-file MiB ceiling. - fireworks: had no entry; its 10 MB budget is on the base64 total, so the raw-byte equivalent is 7.5 MB. Add perRequestMaxBytes for the combined ceilings, enforced before any upload spend, and cover the OpenAI upload path end to end — it had no test at all. * chore(deps): upgrade @google/genai to 2.13.0 and @anthropic-ai/sdk to 0.115.0 @google/genai 2.x reworks the Interactions API, which the Gemini deep-research provider is built on. Migrate it: - `Interaction.outputs` (a flat content array) is now `steps`, a discriminated timeline; the report text lives in the `model_output` steps' text content, alongside thought and tool steps we skip. - `Usage.total_reasoning_tokens` is now `total_thought_tokens`. The old code already fell back to that name through a cast, so this just makes the field the SDK actually returns the typed one. - SSE events renamed: `content.delta` -> `step.delta`, `interaction.start` -> `interaction.created`, `interaction.complete` -> `interaction.completed`. The new event types are discriminated, so the payload casts are gone. Both `interactions.create` calls also stop annotating their params with `Interactions.CreateAgentInteractionParams{,Non}Streaming`. In 2.13.0 those namespace aliases resolve to `CreateAgentInteraction`, whose `stream` is a plain `boolean` rather than a literal — annotating with them erases the discriminant and the call resolves to the union-returning overload, so the result is typed as `Interaction | Stream` at every use. An inline `stream: true as const` keeps the correct overload. Neither upgrade required a `minimum-release-age` waiver: 2.15.0 and 0.115.0 were checked and 2.13.0 is the newest genai release clearing the 7-day window. * chore(deps): upgrade openai to 7.0.0 v5 is the only major with real breaking changes for us; v6 widened a Responses output type and v7 only raised the Node floor to 22, which apps/sim already requires. Three things needed fixing: `ChatCompletionMessageToolCall` became a union of function and custom tool calls, and the custom variant has no `function` field — 43 unguarded `.function` accesses across the OpenAI-compatible providers. Narrow once at each `message.tool_calls` read through a shared `isFunctionToolCall` guard rather than casting at every use. That guard deliberately tests for the `function` payload instead of `type === 'function'`. Many OpenAI-compatible vendors omit `type` on tool calls entirely — our own fixtures do — so discriminating on it type-checks perfectly and then silently drops every tool call those providers return. `ChatCompletionCreateParams.verbosity` narrowed from `string` to a literal union, and the Responses API's output and input item unions now diverge on members Sim never emits (computer-use call outputs, whose `status` admits `failed`, and the `AdditionalTools` escape hatch). Echoing output back as input is what a tool loop is supposed to do, so that conversion is asserted once in convertResponseOutputToInputItems and the streaming loop now routes through it instead of pushing raw output items. The hand-rolled multipart upload in file-attachments.server.ts can now be replaced with the SDK's typed `expires_after` — left for a follow-up so this commit stays a pure upgrade. * fix(providers): correct defects found auditing the attachment and SDK changes The mechanical rewrite that added `isFunctionToolCall` to every `tool_calls` read also rewrote three truthiness guards, where the filtered array was computed, discarded, and the unfiltered value used in the body. Filter once and use that value. The helper also landed between `trackForcedToolUsage`'s TSDoc block and its declaration, leaving that block documenting the wrong function. Raise the Bedrock ceiling from 3.75 MB to 4.5 MB. Converse caps an image at 3.75 MB and a document at 4.5 MB, and a single `maxBytes` cannot express both. Taking the lower bound looked conservative but regressed 3.75-4.5 MB documents, which Converse accepts and which work today. At the document bound every size that works now still works, and only genuinely-too-large files are rejected early; oversized images in that band keep surfacing as a Bedrock API error, exactly as they do without the entry. Both limits re-verified verbatim against the primary docs: Converse's Message reference ("Each image's size ... no more than 3.75 MB", "Each document's size must be no more than 4.5 MB") and Fireworks' vision guide ("Total base64-encoded images must be less than 10MB"). * fix(providers): keep every attachment that works today working Auditing the routing change for backwards compatibility turned up two bands it silently broke. Lowering the single inline cap made the upload path mandatory above ~6 MB, but every large-file path reads its bytes back out of cloud object storage. A deployment without it — local dev, any disk-backed self-host — inlines those files as base64 today and would have started failing outright with "requires cloud file storage". Split the one number in two: the inline ceiling stays at 10 MiB, and a separate threshold marks where an upload becomes *preferable* because the base64 copy no longer fits the payload store. Where no upload path is reachable, base64 hydration now runs to the inline ceiling as before, and a missing cloud-storage backend leaves the file for the inline path instead of throwing. The two strategies also cross over at different sizes now. `files-api` carries every type the provider already accepts, so it takes over at the lower threshold. `remote-url` only fetches images and PDFs, so switching early would have started rejecting 6-10 MB text documents that inline fine today; it takes over only once inlining is genuinely impossible. Revert the Groq ceilings. Its published "20MB" governs a request carrying an image URL, and on this path the body holds only the URL, so it cannot bind on the files maxBytes guards. Groq documents no limit on the image it fetches, so tightening the per-file cap to 20,000,000 and summing raw bytes against the request cap would both reject uploads that work today on no documented basis. * fix(providers): drop the provider ceiling changes and close the audit findings A six-agent line-by-line audit against the vendors' live docs found the `models.ts` ceiling work was not the strict improvement it was written as, so all of it is reverted: - bedrock's 4.5 MB cap broke video. Converse takes image, document AND video blocks, and video is allowed 25 MB base64 — a single `maxBytes` cannot express three content classes, and every 4.5-10 MiB `.mp4` that works today would have started failing. - openai's combined 50 MB cap is the FILE-input limit. Image inputs are governed separately at 512 MB / 1500 images, so summing every attachment rejected eight 8 MB PNGs that OpenAI documents as legal. - fireworks' per-file ceiling was unreachable behind the request budget, while the upload picker went on advertising it — a size the UI accepts and execution always rejects. - The whole `perRequestMaxBytes` feature goes with them: it summed raw bytes against caps that are variously on encoded bytes, on one content class, or on a body that carries only URLs, and it double-counted a file referenced from several messages even though the uploader dedupes by key. Only openai's per-file `maxBytes` stays corrected, to decimal 50,000,000 — the one number a vendor states unambiguously and writes no MiB against. Also fixed, all found by the same audit: The hydration cap stopped short of where `remote-url` actually switches over, so 6-10 MiB attachments on anthropic/openrouter/xai/groq/together/baseten/vllm had neither base64 nor a handle and failed outright — the very band this branch exists to fix. Both decisions now come from one function so they cannot drift. Eight more sites where the mechanical rewrite computed a filtered array and then read the unfiltered one (deepseek, sakana, nvidia, kimi), leaving those providers without the narrowing they appear to have. `isFunctionToolCall` threw on a null or primitive `tool_calls` entry, because `in` requires an object — reachable exactly on the self-hosted gateways this filter was added for. It is now total, and all 32 test mocks match it rather than being quietly more permissive. `checkForForcedToolUsage` in utils/litellm/mistral evaluated the response before the `tool_choice` test, turning a tolerated malformed body into a TypeError on a path that never used to touch it. Gemini: `satisfies` restores the excess-property checking the dropped annotations removed, the poll loop recognises the terminal statuses v2 added instead of spinning for an hour and reporting a timeout, and the streaming doc block no longer names five events that were renamed six lines below it. * fix(providers): report attachment limits in the unit vendors publish The size ceilings are decimal MB — that is how OpenAI, AWS and Fireworks all write them — but the error messages divided by 1024², so OpenAI's 50 MB cap was reported to the user as "48MB". Someone shrinking a 49 MB file to get under it was chasing a limit that does not exist. One formatter, used by all three messages, so the file size and the ceiling in the same sentence are always in the same unit. * fix(providers): derive the limit unit from the ceiling it belongs to The previous commit fixed OpenAI's "48MB" by dividing every ceiling by 10⁶ — which broke the other seven. Only OpenAI's constant is decimal; anthropic, google, together and openrouter are 50 MiB, baseten and vllm 25 MiB, groq and xai 20 MiB. Rendering those as decimal MB overstated each by ~5%, so a 21 MB file on groq was rejected with "(21MB) exceeds the 21MB limit" — a sentence that contradicts itself and sends the user to shrink a file to a size that is still over. Same class of bug as the one being fixed, sign flipped. Both figures now render through one unit taken from the ceiling, so the number a user is told is the number the vendor publishes and the two sizes in a sentence are always comparable. Tested against every ceiling in the registry rather than only the values that happened to round cleanly. Two more from the same audit: A file with a missing or zero declared size was stranded on a files-api provider: hydration bailed on the real byte length while `shouldUseLargeFilePath` saw `0 > threshold` as false, so it got neither base64 nor a handle and failed as "may no longer be accessible" — a size failure wearing an access failure's message. Uploads read the real bytes and enforce the ceiling themselves, so an unknown size now routes to one. The oversized-attachment error blamed the provider for a deployment problem: a files-api provider on a host without cloud storage reported that the provider "has no large-file upload path", which is not true of the provider. `isFunctionToolCall` only proves `function` is present, never that it is well formed, so the trace enricher is defensive again about a hollow payload without giving up the compile-time gate. The 32 test mocks now match production exactly. * fix(providers): stop an over-limit size rendering as the limit itself Deriving the unit from the ceiling fixed the 5% error but left the precision fixed at two decimals, so a file one byte over a 20 MiB cap still printed "(20.00MB) exceeds the 20MB agent attachment limit" — the same self-contradicting sentence, now in a ~5 KB band above every ceiling in the registry. The size rounds up and the ceiling rounds down, so the two can no longer collide. The test that was supposed to guard this asserted a file 0.03MB over and an OpenAI file that was under the limit — neither anywhere near the band — so it passed while the bug was live. It now walks `limit + 1` for every ceiling, and goes red against the old rounding. The reason clause added last commit also claimed a deployment had no cloud file storage whenever the strategy was not inline. A generated document on a remote-url provider reaches that same error with storage fully configured, because a signed URL points at the generation source rather than the rendered artifact — so it was told something false about its own deployment. That case now names itself. * fix(providers): order the attachment failure reason by how general the cause is The generated-document arm was checked first, so it won over both other causes and told users two things that were not true. On an inline-strategy provider — bedrock, mistral, ollama, fireworks, litellm, vertex, kimi — there is no upload path for any file, generated or not, but the message blamed the document format and implied a plain PDF would go through. On openai or google with cloud storage unconfigured it was simply false: a generated document does take the Files API path there, and that exact file uploads fine once storage exists. The one actionable fix was hidden from the operator. A provider with no upload path cannot be helped by changing the file, and a deployment with no object storage cannot reach any upload path whatever the file is, so both now outrank the format-specific case — which is left saying only what is true of it: a signed URL points at the generation source rather than the rendered file. The formatter is unchanged. It was brute-forced over every real ceiling and three million random pairs with no collision or inversion, but the test's six ceilings all divide to exact integers, so floor, round and ceil are indistinguishable on them and the limit-side rounding was unpinned. A ceiling with a fractional remainder now covers it.
…rbosity, and thinking level (simstudioai#6233) * improvement(agent): allow variable references in reasoning effort and verbosity Reasoning Effort and Verbosity were select-only dropdowns, so a workflow could not sweep them from a variable or an upstream block the way it already can with the model. Both become editable comboboxes, matching the model field directly above them and the managed-agent selectors. - switch both subblocks to `combobox`, keeping their fetched per-model option lists intact - keep them visible when `model` itself holds a reference, since the concrete model id is only known at execution time and cannot be matched against the static capability list - normalize the resolved level in the provider chokepoint so a reference that resolves to `"High"` or to nothing behaves sanely instead of hitting a provider 400 * improvement(agent): allow variable references in thinking level Extends the same treatment to Thinking Level so all three model-tuning fields behave consistently, and logs a level a model does not declare. - switch `thinkingLevel` to `combobox` with the reference-aware condition - normalize it alongside the other two; an empty resolve now takes the deliberate "send nothing" path rather than the incoherent half-state it hit before, and stays distinct from an explicit `none` - warn when a level is not one the model declares, still forwarding it: Sim's per-model lists drive the pickers and can lag a provider, and a sweep needs the provider's own error rather than a silent fallback to the default * improvement(agent): report a model level the sanitizer discards A model bound to a variable or block reference only resolves at execution time, so a run whose reference landed on a model outside Sim's catalogue had its requested level cleared with no signal and quietly fell back to that model's default. Dropping stays the safe default — a provider with no such parameter rejects the whole request — but it is now reported. - log the field, model, and value whenever an unsupported-field level is cleared - cover both diagnostics, including that they stay quiet for a declared level and for the `auto` / `none` sentinels * fix(providers): redact resolved level content from sanitizer diagnostics The model-level fields accept environment and block references, so an unrecognized level is not necessarily a mistyped level — it is whatever the reference resolved to, which can be secret content. The diagnostics added for dropped and undeclared levels echoed it straight into server logs. - log a level only when the catalogue declares it somewhere, or it is an `auto` / `none` sentinel; anything else is reported by length alone - stop discarding levels for a model the catalogue has never seen. Absent is unknown, not known-incapable, and a reference is exactly how a newly released model arrives before Sim catalogues it — the provider decides instead. Models the catalogue knows, and every dynamic-provider id, keep the protective drop * fix(providers): redact the level in Anthropic's unsupported-thinking warning Forwarding an undeclared level is deliberate, but it means the Anthropic adapter receives it and interpolates it straight into its "not supported, ignoring" warning. Since the field is reference-bound, that value can be whatever a mistyped `{{ENV_VAR}}` or block reference resolved to — so the redaction added for the sanitizer's own diagnostics was leaking one layer downstream. - promote the level renderer to `providers/utils` as `describeModelLevel`, the single gate every site echoing a caller-supplied level goes through - use it in Anthropic's warning and in both sanitizer diagnostics * refactor(providers): drop the sanitizer's level diagnostics The two warnings logged server-side, where the workflow author who set the level never sees them, and the surprising case they described — a level discarded for a model newer than the catalogue — is now fixed at the source rather than narrated. They also carried the redaction that leaked resolved content before it was caught, so removing them removes that surface entirely. Levels still normalize, and still drop for a catalogued model that does not take the field. `describeModelLevel` stays for Anthropic's unsupported-thinking warning, which is a pre-existing log this feature newly exposes to resolved reference content.
* fix(docker): upgrade bun to 1.3.14 to unbreak the Next 16.3.0 server
Bun 1.3.13 cannot load Next 16.3.0's compiled server runtime. The app container
runs the Next server under Bun (`oven/bun:1.3.13-slim`, `bun apps/sim/bootstrap.js`),
so every app-page render threw and `/api/health` returned 500:
⨯ Error: Failed to load external module
next/dist/compiled/next-server/app-page-turbo.runtime.prod.js:
TypeError: Expected CommonJS module to have a function wrapper.
If you weren't messing around with Bun's internals, this is a bug in Bun
Isolated to Bun, not Next, by loading that exact module in the real images:
Next 16.2.12 + Bun 1.3.13 -> loads (why staging was fine before)
Next 16.3.0 + Bun 1.3.13 -> CJS wrapper error
Next 16.3.0 + Bun 1.3.14 -> loads
Bun 1.3.14 is the current stable and already fixes it, so this bumps every pin
rather than reverting the framework upgrade, which would only defer the same
latent Bun bug to the next attempt.
Why no gate caught it: local dev machines and this bump's own verification run
Bun 1.3.14, while the container and CI pinned 1.3.13 — and CI only *builds* the
image, it never boots one and probes `/api/health`. A container smoke test in CI
would have caught this before merge; that is worth adding separately.
* fix(docker): align the remaining bun pins with 1.3.14
Two pins were missed in the first pass because the search was scoped to
docker/, package.json and .github/workflows/:
- .devcontainer/Dockerfile still built on oven/bun:1.3.13-alpine
- PI_BUN_VERSION in apps/sim/scripts/pi-sandbox-packages.ts was still 1.3.13,
despite being documented as mirroring the root packageManager field, so Pi
sandbox images would have kept installing the Bun release that cannot load
the Next 16.3.0 server runtime.
Fixed surgically rather than with a repo-wide replace: "1.3.13" also appears
inside SVG path data in apps/sim/components/icons.tsx and
apps/docs/components/icons.tsx, which a blind sed would have corrupted.
…ocs overhaul, nextjs and bun upgrade
…code (simstudioai#6242) Next 16.3.0's Turbopack optimizer models a bare `return <asyncCall>()` tail call inside an async function as returning the promise object, then propagates that always-truthy fact through the caller's `await`. Where the result feeds an `if (x)` whose every branch returns, it concludes the branch is always taken and deletes everything after it from the emitted bundle. Two sites shipped to production that way: - `POST /api/credentials` lost its entire create path — the transaction, the org locks, the insert, the audit, the 201. A first-time create fell into the existing-credential branch and threw on `existingCredential.id`, so every new credential 500'd. - `upsertAsyncToolCall` collapsed to `async () => await getAsyncToolCall(id)`. The insert is simply gone; it returns null for every new async copilot tool call. Silent — no error, no failed request. A differential scan of 71,266 source string literals across `.next/server` and `.next/static`, comparing images built from the same commit on 16.2.12 and 16.3.0, found exactly these two and nothing else. That scan cannot see dropped branches with no distinctive string literal, which is why the version goes back rather than the two sites being patched alone. Both are also hardened with `return await`, verified to defeat the miscompile in a minimal reproduction. The TypeScript toolchain cleanup from the original bump (dropping @typescript/native-preview, `useTypeScriptCli`) is kept.
Collaborator
Author
Open questions — upstream sync 2026-08-04Two blocking decisions. Everything else in this 518-commit range is resolved from Reply on this PR with Q1 — 🔴 Six upstream migrations will be silently skipped on every existing Arena database. How do we fix it?This is the most serious finding of the run. It is not a merge conflict, it does not fail Why. select id, hash, created_at from drizzle.__drizzle_migrations order by created_at desc limit 1then applies a file only when The fork and upstream both used idx 0258–0261, and their journal
What never gets created: the Aggravating factor: fork migrations Options(A) Re-stamp both sides — recommended.
(B) One-off
(C) One-shot script migration. Use the existing
→ Which option? (Default if you have no preference: A.) Q2 — 🔴 Upstream deleted the deployed-chat voice/TTS stack. Does Arena keep voice mode?Upstream removed voice mode wholesale (simstudioai#6215, simstudioai#6218, simstudioai#6220, simstudioai#6224), deleting:
The fork's fork-only
Options(A) Keep voice mode. Skip the seven deletions for these paths, keep the fork's voice stack, (B) Drop voice mode, following upstream. Accept all seven deletions and strip voice-first → (A) or (B)? If (A), please also confirm whether voice should stay on the deployed/public Not asking about these — recorded as decisions in
|
Collaborator
Author
|
/upstream-sync resume
|
Collaborator
Author
|
/upstream-sync resume |
check: secrets rejects replica-host as a remote credential URL. Co-authored-by: Cursor <cursoragent@cursor.com>
Collaborator
Author
|
Closing to retry upstream-sync with the release-sliced merge-plan harness on feat/github-merge-agent. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
8 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Upstream sync — 2026-08-04
Merges
simstudioai/sim@1b9e0f25intofeat/github-merge-agent.Sync range: 518 commit(s) since
e2fecc86(merge-base).Ledger
Verification (advisory — does not block the sync)
✅
bun run check· ❌bun run lint· ❌bun run test· ❌bun run buildVerification is advisory — failures do not block the sync. Review and fix on the draft PR as needed.
bun run check
✅ passed
bun run lint
❌ failed
bun run test
❌ failed
bun run build
❌ failed
Agent usage
child-cluster-1
gpt-5.6-lunaTotals
Cost by agent
Draft — mark ready for review when satisfied.