feat: unify server-side selector execution

[BillLeoutsakosvl346]

Summary

Moves every dynamic selector onto one exhaustive, server-authorized execution architecture while preserving the existing dropdown/combobox UX, React Query behavior, pagination, search, detail hydration, and saved workflow/connector formats.

• Classifies all 93 dynamic selector keys in one browser-safe manifest:
  • 81 provider-server
  • 11 internal-server
  • 1 local (workspace.triggerTypes)
• Executes all 92 server selectors through the selectors.execute application operation and POST /api/selectors/execute.
• Resolves exact {{KEY}} references only after session and canonical workflow/workspace authorization.
• Authorizes credential use and provider/service binding before provider work.
• Returns only explicitly projected selector options; resolved context, tokens, credentials, and raw provider responses never cross the boundary.
• Migrates workflow/block/trigger, connector, fork, search/replace, comparison, dynamic-option, and display-hydration consumers.
• Deletes the complete browser provider-selector registry and the remaining 70 obsolete selector routes. The two CloudWatch listing routes and Confluence /page remain as thin compatibility adapters over staging's in-process operations; selectors do not call them.
• Adds focused IMAP policy parity across selection, deployment validation, and polling.
• Adds add-selector / validate-selector skills and updates adjacent integration, connector, trigger, block, and application-operation guidance.

There is no feature flag, browser-side provider fallback, provider-legacy classification, database migration, or saved-configuration migration.

Architecture

Before

flowchart TD
  A["Workflow, trigger, connector, fork, search, comparison, display"]
  --> B{"Surface-specific context builder"}
  B --> C["Browser personal-environment resolution"]
  B --> D["Unresolved dependency values"]
  C --> E["Client provider registry and provider-specific query key"]
  D --> E
  C --> X["Hidden shared value unavailable: dependency omitted or disabled"]
  E --> F["Provider-specific API route"]
  E --> G["Browser OAuth-token helper for some providers"]
  G --> F
  F --> H["External provider"]

Loading

After

flowchart TD
  A["Workflow, trigger, connector, fork, search, comparison, display"]
  --> B["One active-value context builder: literals plus exact unresolved references"]
  B --> C["One React Query facade with opaque privacy-safe identity"]
  C --> D["POST /api/selectors/execute"]
  D --> E["Authenticate session and resolve canonical scope"]
  E --> F["Authorize selectors.execute"]
  F --> G["Manifest capability and exact context allowlist"]
  G --> H["Resolve exact references server-side"]
  H --> I["Authorize credential use and provider/service binding"]
  I --> J["Apply destination policy and execute trusted adapter"]
  J --> K["Project, bound, sanitize, and return options only"]
  B --> L{"Local classification?"}
  L -->|"workspace.triggerTypes only"| M["Pure local attachment"]

Loading

Security boundary

• Session authentication occurs before request-body parsing.
• Workflow scope derives its workspace canonically; a supplied workspace ID is assertion-only and mismatch is concealed.
• Selector keys, capabilities, context fields, credential services, provider scopes, operation kinds, and destinations are code-defined—not client-selected.
• Only full-string environment references are resolved. Embedded interpolation is rejected.
• Effective environment precedence remains workspace over personal; missing and inaccessible references share the same public error.
• Credential existence, authorization, workspace membership, and provider mismatch share the same public connection error.
• Query keys contain no context value, reference, credential ID, raw secret, resolved value, search/detail value, or value-derived hash.
• The selector route is private, no-store; there is no selector result cache or high-volume selector audit event.
• Protected values are tracked through execution and fail closed if projected option data contains protected plaintext.
• Provider/SDK failures are caught and reduced to small typed public errors.
• IMAP preserves literal/manual and viewable-reference behavior, but rejects hidden use-only username/password references before DNS. Host validation is DNS-pinned, TLS or mandatory STARTTLS is required, deployment/polling resolve just in time, and sensitive host validation suppresses hostname/IP/raw resolver details in logs.

Authorized use of a shared credential intentionally declassifies only the provider resource metadata represented by normalized options; it does not declassify the credential or referenced secret itself.

Explicit process-local token-cache exception

Selector data never enters browser state, query/cache keys, selector result caches, responses, logs, errors, audit metadata, or rate-limit keys. Authorized client-credential resolution may reuse the shared process-local token cache after authorization and provider binding.

This exception applies only to short-lived client-credential bearer tokens minted by the existing credential service. In selector privacy mode, credential-derived cache identities are HMACed. Each resolution reauthorizes the request, verifies provider binding, rereads the credential row, and validates the current encrypted-secret fingerprint before using a cached token, so credential rotation or reconnection prevents stale-token reuse. Entries remain in one server process, become ineligible for reuse at token expiry, are removed lazily on a later client-credential resolution or process exit, and never cross the selector boundary.

Explicit security-owner acceptance of this exception is required before merge. If it is not accepted, selector privacy mode must bypass client-credential token-cache reads and writes.

Migration inventory

• 93 manifest keys = 92 exhaustive server attachments + 1 exhaustive local attachment.
• 187 workflow/block/trigger declarations and 32 connector fields resolve to registered keys.
• 33 browser provider modules and the shared browser selector infrastructure are removed.
• 70 selector-only provider/API routes remain in the current diff and are removed; exact literal URL auditing found no deleted-route consumer.
• CloudWatch log-group/log-stream routes and the Confluence page route remain as thin compatibility adapters over staging's in-process operations; selector adapters use their stricter server-only provider primitives directly.
• lib/api/contracts/selectors now contains only the unified execute contract; retained CloudWatch, Confluence, and JSM tool contracts live under contracts/tools, and the Harmonic provider schema is adapter-local.
• Diff against current staging: 330 files, 13,696 additions, 20,382 deletions.

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation
• Other: security architecture migration

Testing

Focused migration suites:

• 20 test files, 191 tests passing across the route, executor, manifest, context builder, client facade, representative adapters, workflow search/display, OAuth privacy, and IMAP runtime policy.
• Independent IMAP/security re-audit: 6 suites, 440 tests passing.
• Independent completeness audit: 93/92/1 counts reconciled and every deleted route checked for literal consumers.
• Independent client/UX audit: active registry-backed search, credential aliases, impersonation, and privacy-safe query revisions verified.
• Review and staging-merge regressions: 20 focused files, 210 tests passing across environment ACLs, route concealment, context projection, query invalidation, provider binding, exact-reference sanitization, and retained CloudWatch/Confluence/JSM operation compatibility.
• Final reference-hardening follow-up: 5 focused files, 39 tests passing across exact-reference collisions, requested-key environment resolution, and the shared testing mock.
• Full repository run: all 19 workspace tasks passed; the Sim app completed 2,638 passing test files and 36,509 passing tests.

Repository gates:

• bun run --cwd apps/sim type-check
• bun run --cwd packages/testing type-check
• bun run check:audits (all 37 audits)
• bun run check:route-verbs
• bun run check:api-validation:strict
• bun run check:fork-dependent-coverage
• bun run check:react-query
• bun run check:client-boundary
• bun run check:tool-registry-boundary
• bun run check:skills
• bun run lint:check (all 26 packages pass without warnings)
• bun run --cwd apps/sim build
• git diff --check

Review follow-up fixes

• Active dependency projection now ignores dormant, noncanonical fields; Slack OAuth and bot-token mode switches cannot override one another.
• Missing, mismatched, and inaccessible workflow/workspace scopes now share the same private, non-cacheable 404 response.
• workspace.secretNames uses an ACL-aware names-only environment path that does not decrypt values or populate the decrypted snapshot cache.
• OneDrive folder and SharePoint site selectors accept the exact Word/Excel credential aliases their blocks use, while file/list selectors remain narrowly bound.
• Exact-reference detail hydration now sanitizes provider data with a detail-scoped exemption only for proven exact-ID repeats, then restores those repeats to the browser-owned {{KEY}}. The ID key / ID value collision is covered; partial or unrelated protected plaintext still fails closed.
• Exact selector references now use a fresh ACL-aware requested-key resolver that decrypts only unique referenced values and neither reads nor warms the runtime snapshot cache.
• Environment and credential mutations invalidate both normal selector queries and workflow search/replace selector queries through one privacy-safe helper.

Known limitation and deferred follow-ups

The review also identified adjacent pre-existing/shared concerns that are intentionally outside this PR’s locked scope:

• Excel-backed SharePoint site discovery remains a known limitation: Microsoft Graph requires Sites.Read.All, while current Excel consent is file-only, so ordinary Excel credentials are expected to receive 403. The consent and credential model is tracked in #7197; no OAuth-scope or reconsent change is included here.
• Thread authoritative custom and historical snapshot schemas through registry-independent selector context building.
• Extend end-to-end cancellation into shared token minting and the OpenRouter catalog without breaking shared singleflight work.

Reviewer focus

1. Authorization and resolution ordering in the central executor.
2. Manifest/context readiness equivalence across workflow and connector surfaces.
3. Credential provider binding and option projection in special adapters (Gmail/service accounts, Atlassian, Slack, Microsoft, Zoho, Snowflake/NetSuite, Managed Agent, CloudWatch).
4. IMAP's deliberate hidden-auth restriction and deployment-actor behavior.
5. Selector-only route deletion versus the retained runtime-shared routes.

Deferred live browser validation

The PR remains draft until disposable provider credentials are available. Automated tests do not require live credentials.

• Gmail OAuth plus advanced credential-ID environment reference.
• Jira credential plus domain environment reference.
• CloudWatch raw access-key/secret/region references.
• IMAP literal/manual success and hidden-use-only auth rejection.
• One connector cascade (Google Drive or Airtable).
• Optional search/pagination-heavy provider (Slack or Microsoft).

Excel-backed SharePoint site discovery is not an acceptance item for this PR and is expected to fail until #7197 selects and implements a consent model. Native SharePoint credential validation remains part of any Microsoft live-provider coverage.

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• Live-provider browser checklist completed
• Security owner accepts the process-local client-credential token-cache exception and its lazy-eviction behavior
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

Screenshots/Videos

No visual UI redesign. The selector UX is intentionally preserved; live-provider screenshots are deferred with the credential checklist above.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Aug 28, 2026 11:59pm

[greptile-apps]

Too many files changed for review (296 files, 100 file limit).

Bypass the limit by tagging @greptile-apps to review.

[cubic-dev-ai]

All reported issues were addressed across 296 files

_{You’re at about 90% of the monthly reviewed-line limit. You may want to disable incremental reviews to conserve quota. Reviews will continue until that limit is exceeded. If you need help avoiding interruptions, please contact contact@cubic.dev.}

<sub>Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.

Re-trigger cubic</sub>

[BillLeoutsakosvl346]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the addressed selector, OAuth, IMAP, provider-boundary, and UX findings; report only remaining or newly introduced issues.

[cubic-dev-ai]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the addressed selector, OAuth, IMAP, provider-boundary, and UX findings; report only remaining or newly introduced issues.

@BillLeoutsakosvl346 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

All reported issues were addressed across 314 files

<sub>Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.
Tip: instead of fixing issues one by one fix them all with cubic

Re-trigger cubic</sub>

[BillLeoutsakosvl346]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the addressed selector, OAuth, IMAP, provider-boundary, and UX findings; report only remaining or newly introduced issues.

[cubic-dev-ai]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the addressed selector, OAuth, IMAP, provider-boundary, and UX findings; report only remaining or newly introduced issues.

@BillLeoutsakosvl346 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

All reported issues were addressed across 315 files

<sub>Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.

Fix all with cubic | Re-trigger cubic</sub>

[BillLeoutsakosvl346]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the repaired existing-IMAP deployment binding, connector selector validation guidance, and the documented Confluence and NetSuite dispositions. Report only remaining or newly introduced issues.

[cubic-dev-ai]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the repaired existing-IMAP deployment binding, connector selector validation guidance, and the documented Confluence and NetSuite dispositions. Report only remaining or newly introduced issues.

@BillLeoutsakosvl346 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

All reported issues were addressed across 315 files

<sub>Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.
You've manually re-run cubic several times on this PR. Each manual re-review checks the full PR again and counts toward your usage quota. To preserve your usage limits, we recommend letting cubic automatically review new commits.

Fix all with cubic | Re-trigger cubic</sub>

[BillLeoutsakosvl346]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the existing-IMAP deployment binding update and rollback paths, including failed polling setup. Report only remaining or newly introduced issues.

[cubic-dev-ai]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the existing-IMAP deployment binding update and rollback paths, including failed polling setup. Report only remaining or newly introduced issues.

@BillLeoutsakosvl346 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

All reported issues were addressed

<sub>Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.
You've manually re-run cubic several times on this PR. Each manual re-review checks the full PR again and counts toward your usage quota. To preserve your usage limits, we recommend letting cubic automatically review new commits.

Fix all with cubic | Re-trigger cubic</sub>

[BillLeoutsakosvl346]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check all addressed selector, OAuth, IMAP, provider-boundary, and UX findings, including existing-IMAP deployment binding update and rollback paths. Report only remaining or newly introduced issues.

[cubic-dev-ai]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check all addressed selector, OAuth, IMAP, provider-boundary, and UX findings, including existing-IMAP deployment binding update and rollback paths. Report only remaining or newly introduced issues.

@BillLeoutsakosvl346 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

No issues found across 315 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

<sub>Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.
You've manually re-run cubic several times on this PR. Each manual re-review checks the full PR again and counts toward your usage quota. To preserve your usage limits, we recommend letting cubic automatically review new commits.

Re-trigger cubic</sub>

[BillLeoutsakosvl346]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the IMAP active-deployment webhook lookup and rollback paths together with all previously addressed selector, OAuth, provider-boundary, and UX findings. Report only remaining or newly introduced issues.

[cubic-dev-ai]

@cubic-dev-ai rerun and review the complete updated PR against staging. Re-check the IMAP active-deployment webhook lookup and rollback paths together with all previously addressed selector, OAuth, provider-boundary, and UX findings. Report only remaining or newly introduced issues.

@BillLeoutsakosvl346 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

No issues found across 315 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

<sub>Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.
You've manually re-run cubic several times on this PR. Each manual re-review checks the full PR again and counts toward your usage quota. To preserve your usage limits, we recommend letting cubic automatically review new commits.

Re-trigger cubic</sub>