Refresh the PR onto current main and close the remaining probe-setting review gaps: the primary site save now persists the latency threshold, the site row type carries that field without an any-cast, active probe streams are aborted on unmount, and all-model one-shot probes get the longer timeout they need.
Constraint: PR #510 must preserve the per-site model probe feature while merging cleanly into current main.
Rejected: Leave threshold persistence behind the separate probe-settings button | the primary save path would still silently drop user changes.
Rejected: Keep the SiteRow any-cast | it hides API/schema drift for the new persisted field.
Confidence: high
Scope-risk: narrow
Directive: Keep probe setting fields represented in the typed site save payload and covered by focused web API/editor tests.
Tested: npm run typecheck; npm run repo:drift-check; npx vitest run --root . src/server/db/schemaContract.test.ts src/server/db/schemaArtifactGenerator.test.ts src/server/db/schemaIntrospection.test.ts src/server/db/schemaParity.test.ts src/web/api.test.ts src/web/pages/helpers/sitesEditor.test.ts src/web/pages/sites.disabled-models-save.test.tsx src/web/pages/sites.edit-scroll.test.tsx
Not-tested: Full npm test.
The release branch is main-based and tracks the reviewed dev slice for PR publication. This merge carries the post-review compatibility fixes without rewriting the already-open PR history.
Constraint: PR #533 publishes dev changes through a clean release branch targeting main.
Rejected: Recreate the PR branch from scratch | unnecessary churn after green checks and review context already exist.
Confidence: high
Scope-risk: narrow
Tested: npx vitest run --root . src/server/proxy-core/responsesPreflight.test.ts src/server/proxy-core/serviceTierPolicy.test.ts src/server/routes/proxy/chat.stream.test.ts src/web/pages/settings/UpdateCenterSection.test.tsx
Tested: npm run typecheck:server
Tested: npm run typecheck:web
Tested: npm run repo:drift-check
Co-authored-by: OmX <omx@oh-my-codex.dev>
Review feedback found narrow compatibility regressions in the web-search shim, Responses tool detection, service tier accounting, and update-center Docker candidate actions. The fix keeps synthetic search routed through the dedicated search model, preserves Responses text compatibility, distinguishes absent service_tier from filtered service_tier, and evaluates recent Docker tags per candidate.
Constraint: Review fixes must stay on dev first, then flow into the main-based release PR branch.
Rejected: Leave Responses web_search simulation on the request model | it has the same search-route selection risk as Anthropic simulation.
Confidence: high
Scope-risk: narrow
Tested: npx vitest run --root . src/server/proxy-core/responsesPreflight.test.ts src/server/proxy-core/serviceTierPolicy.test.ts src/server/routes/proxy/chat.stream.test.ts src/web/pages/settings/UpdateCenterSection.test.tsx
Tested: npm run typecheck:server
Tested: npm run typecheck:web
Tested: npm run repo:drift-check
Co-authored-by: OmX <omx@oh-my-codex.dev>
Several downstream clients send non-cache compatibility fields across Chat, Responses, Anthropic Messages, and Responses WebSocket surfaces. The proxy now normalizes legacy functions/function_call, maps web_search tools across protocols, rejects unsupported HTTP Responses continuation shapes before upstream dispatch, and applies configurable service_tier policy consistently before requests leave the server.
Constraint: Publication flow keeps day-to-day commits on dev and opens main PRs from a release branch.
Constraint: No new dependencies; web_search-only simulation reuses the existing local search route.
Rejected: Forward external HTTP previous_response_id to strict upstreams | it produces ambiguous continuation failures and stale retry behavior outside WebSocket state.
Rejected: Add a dedicated search dependency | existing /v1/search route is sufficient for synthetic server-tool responses.
Confidence: high
Scope-risk: moderate
Directive: Keep Responses HTTP continuation checks aligned with WebSocket continuation support before relaxing these guards.
Tested: npm run typecheck:server
Tested: npx vitest run --root . src/server/proxy-core/responsesPreflight.test.ts src/server/proxy-core/serviceTierPolicy.test.ts src/server/test-fixtures/protocol/non-cache-field-matrix.test.ts src/server/transformers/openai/responses/conversion.test.ts src/server/transformers/anthropic/messages/conversion.test.ts src/server/routes/proxy/chat.stream.test.ts src/server/routes/proxy/responses.codex-oauth.test.ts
Tested: npx vitest run --root . src/server/routes/proxy/responses.websocket.test.ts
Tested: npm run repo:drift-check
Not-tested: Full application build and full Vitest suite
Co-authored-by: OmX <omx@oh-my-codex.dev>
MySQL information_schema.COLUMNS.COLUMN_DEFAULT stores DEFAULT '' as an
empty string, not NULL. normalizeDefaultValueForColumn was treating any
empty rawDefaultValue as 'no default' and returning null, causing the
MySQL schema parity test to see defaultValue: null for post_refresh_probe_model
while schemaContract.json (generated from TypeScript schema) had "''".
Fix: when rawDefaultValue is empty string, return "''" for text/json columns
instead of null, so the round-trip through MySQL introspection matches the
contract.
Better-SQLite3 requires --> statement-breakpoint between each statement
in a migration file; without it all statements are passed as a single
string to Database.prepare() which throws RangeError on multi-statement input.
- Fix SSE abort: propagate AbortSignal to worker pool; close connection
immediately on client disconnect without sending 'complete' event
- Fix Boolean coercion: use strict === true/=== 1 check instead of Boolean()
to avoid Boolean('false') === true foot-gun
- Fix probe-now: parse and forward latencyThresholdMs from request body
- Fix api.ts probeSiteNow: add latencyThresholdMs to options type
- Fix backup import: restore postRefreshProbe* fields when inserting sites
so that backup/restore round-trips preserve per-site probe config
- Fix latency threshold persistence: add postRefreshProbeLatencyThresholdMs
column (migration 0026), save/load it in site editor, pass to
runPostRefreshProbeIfEnabled so auto-probe also respects the threshold
- Fix state management: setProbeCompleted(false) at probe start; error
branch also refreshes model lists and sets probeCompleted to show results
- Regenerate schema artifacts (mysql/postgres bootstrap+upgrade, contract)
- Move post-refresh probe settings from global config to per-site (sites table)
- Add manual 'probe now' button in site editor with real-time SSE log stream
- Use fetch+ReadableStream instead of EventSource to support Authorization header
- Concurrent probe execution (default 10 workers) via worker-pool pattern
- Add stop button to cancel in-flight probe while preserving completed results
- Add latency threshold: auto-disable models exceeding response time limit
- Treat inconclusive probe results as unsupported (auto-disable)
- Show specific failure reason in logs (timeout, no token, model not found, etc.)
- Refresh and display post-probe model status list (green=available, red=disabled)
- Remove probe settings from global Settings page and RuntimeSettingsPayload
The update center already allowed manual Docker tag deploys, but operators
still had to know and paste dev or branch tags by hand. This change keeps the
stable Docker Hub candidate as the main path while automatically surfacing the
most recent non-stable tags and their digests for one-click deployment.
Constraint: Stable Docker Hub discovery must remain the default candidate path
Constraint: Existing deploy API and helper protocol should stay unchanged
Rejected: Promote dev tags into the primary stable candidate slot | would blur stable and experimental release semantics
Rejected: Add a brand new backend deploy endpoint for recent tags | existing targetTag/targetDigest deploy path already covers the need
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep stable and recent non-stable Docker tag discovery distinct unless product semantics deliberately change
Tested: Focused vitest for update-center services/routes/UI, npm run typecheck, npm run repo:drift-check, npm run build:web, npm run docs:build, git diff --check
Not-tested: Full npm test remains red because tmp/sub2api frontend suites are still broken in this repo snapshot
The last CI failure after the review cleanup was only a stale test expectation:
function_call fallback items now correctly use fc_* item ids while preserving the
call_* id in call_id. Updating the test keeps the suite aligned with the
Responses payload shape we intentionally adopted in the bridge.
Constraint: Keep the fix to the assertion only; the production behavior already matches the reviewed contract
Rejected: Revert the fc_* item id change | that would reopen the review thread we intentionally fixed
Confidence: high
Scope-risk: narrow
Directive: When bridge output semantics change intentionally, update sibling facade tests in the same PR so CI reflects the new contract immediately
Tested: Focused vitest for openai responses outbound/response bridge, Gemini request bridge, Settings payload rules; typecheck; git diff --check
Not-tested: Full npm test
The remaining review threads were down to route alias hygiene, proxy-safe header
ordering, and a Gemini canonical tool-result mismatch. This commit makes the
alias/websocket behavior explicit, preserves contributor helper safety, and
teaches the canonical Gemini path to recover tool names and wrap function
results the same way the OpenAI-shaped path already does.
Constraint: Keep the last review fixes limited to the current branch so the release PR stays easy to reason about
Rejected: Defer the Gemini canonical fix to a later refactor | the review found a real functional mismatch and the fix stayed localized to the request bridge
Confidence: high
Scope-risk: narrow
Directive: Keep canonical-to-provider bridges internally consistent; if one path resolves tool names or wraps results, the sibling path should not diverge silently
Tested: Focused vitest for Gemini request bridge, route aliases, upstream request builder, responses response bridge, contributor updater; typecheck; git diff --check
Not-tested: Full npm test; any broader end-to-end Gemini live-provider behavior outside the covered bridge tests
The latest review pass found a few more concrete edge-cases that were still
cheap to close: invalid /responses aliases could still rewrite typos, websocket
session headers should prefer proxy-safe dashed forms, contributor rendering
needed an input guard, and several upstream request/response helpers still had
small contract mismatches around Accept/beta handling and synthesized tool ids.
These fixes keep the release branch behavior explicit while the PR remains open.
Constraint: Keep changes surgical and well-covered so the review cleanup does not destabilize the already-green release branch
Rejected: Leave these for post-merge cleanup | they were still active review threads with straightforward fixes
Confidence: high
Scope-risk: narrow
Directive: When alias routes exist only for ergonomics, reject unknown subpaths loudly instead of silently rewriting them
Tested: Focused vitest for contributor updater, route aliases, upstream request builder, openai responses response bridge; typecheck; git diff --check
Not-tested: Full npm test; the remaining Gemini review thread that likely needs a broader canonical-tool-result contract decision
A few review threads still pointed at real behavior mismatches: Anthropic tool
JSON deltas were still trimming whitespace, Responses conversion still raised
low token caps to the floor, and the bare /responses GET alias returned upgrade
messaging that did not match the actual websocket entrypoint. This commit makes
those behaviors explicit and locks them with focused tests.
Constraint: Keep the release branch green while fixing only the review findings that have a clear, testable contract
Rejected: Leave these for a follow-up cleanup PR | they were low-diff and directly blocked comment resolution on the current release PR
Confidence: high
Scope-risk: narrow
Directive: Preserve raw stream delta text unless a downstream protocol explicitly requires normalization; whitespace can be semantic in incremental payloads
Tested: Focused vitest for anthropic stream bridge, responses conversion, route aliases; typecheck; git diff --check
Not-tested: Full npm test; the remaining review threads that still need broader product decisions
The PR still had a few targeted review threads around contributor pagination,
route-layer passthroughs, dead header fallbacks, and synthetic response ids.
These are small enough to settle directly on the release branch without
broadening the merge scope, and they tighten the repo back toward the
service-owned boundaries this refactor is aiming for.
Constraint: Keep the changes narrow and regression-covered so the release PR stays green while review threads are closed out
Rejected: Leave the minor threads for follow-up after merge | they were isolated enough to fix now and would keep noise on the PR
Confidence: high
Scope-risk: narrow
Directive: When a route shim stops serving any route-local behavior, point callers at the owning service instead of keeping a passthrough around
Tested: Focused vitest for contributor updater, openai responses response bridge, upstream request builder, stats proxy logs, architecture boundaries; typecheck; repo drift check; git diff --check
Not-tested: Full npm test; unresolved review threads that still require larger product decisions
The release branch still had unresolved review comments around payload-rule
atomicity, brand blocking, compact fallback controls, and response
compatibility edge-cases. This patch pulls those behaviors back in line with
the intended contracts and adds focused regression coverage for the fragile
bridges.
Constraint: The fixes must stay narrow to the PR branch so review/merge can continue without reopening the larger dev snapshot conflict set
Rejected: Leave outdated-looking threads unresolved until after merge | reviewers still see concrete unfixed regressions in the current release branch
Rejected: Re-run another broad merge from dev/main | too risky while the PR branch is already green and mergeable
Confidence: medium
Scope-risk: moderate
Directive: If these release-branch fixes need to persist for future release cuts, backport them onto dev instead of re-fighting the same review threads next time
Tested: Focused vitest for settings events, brand matcher, compact responses, codex compatibility, anthropic stream bridge, openai chat response bridge; typecheck; repo drift check; git diff --check
Not-tested: Full npm test; live PR thread resolution state after push
The release branch inherited a few merge-shaped regressions after the
large dev snapshot was layered onto the latest main snapshot. This
commit restores the main-compatible stats and TokenRoutes behavior the
CI suite expects and adds the missing multi-brand helper export so the
release branch reflects a coherent post-merge tree instead of a
partially regressed hybrid.
Constraint: Must keep the release branch mergeable to main while preserving the current publish snapshot shape
Rejected: Re-open the whole release merge and hand-resolve every file again | too broad for targeted CI repair
Confidence: medium
Scope-risk: narrow
Reversibility: clean
Directive: When dev and main have drifted heavily, re-run the main-facing CI tests after any conflict normalization pass
Tested: targeted vitest suite for failing CI jobs plus update-center coverage; typecheck
Not-tested: Full npm test still not representative locally because of unrelated untracked tmp/sub2api suites outside the PR branch content
This release branch starts from the latest origin/main and folds in the
current origin/dev snapshot so the PR to main can follow the repo's
publish flow without opening directly from dev. Conflict-heavy files
were normalized back to the dev snapshot where necessary so the release
branch reflects the code currently being carried on dev instead of a
line-merged hybrid.
Constraint: Publication flow requires a clean release branch from latest origin/main rather than a PR directly from dev
Constraint: The branch must preserve the current dev snapshot even when main and dev have drifted substantially
Rejected: Open the PR directly from dev | violates the repo publication flow
Rejected: Manually redesign every historic dev/main conflict during release prep | too broad for a publish branch
Confidence: medium
Scope-risk: moderate
Reversibility: clean
Directive: Treat this branch as a publication snapshot of dev-on-top-of-main, not as a general cleanup pass over old divergence
Tested: update-center focused vitest suite on release branch; typecheck; build:web; docs:build; repo:drift-check
Not-tested: Full npm test remains red from inherited tmp/sub2api frontend suites and an existing firstByteTimeout unhandled error
The transferred dev snapshot was not publication-ready yet: full tests still
failed because worktree-local docs resolution assumed a repo-root node_modules
tree, some protocol expectations no longer matched the refactored request
bridges, oauth/account-token route tests reused stale modules after resetting
DATA_DIR, and model discovery still fired unawaited default-token writes.
This follow-up makes the moved dev branch self-consistent and green again by
stabilizing module/test isolation, awaiting token mirroring writes, and
aligning narrow tests with the current transformer and request-builder
contracts.
Constraint: The fix set had to stay narrow and unblock the new dev branch without reworking the broader transferred refactor
Rejected: Revert the transferred snapshot | would lose the user's requested dev-branch continuation point
Confidence: medium
Scope-risk: moderate
Reversibility: clean
Directive: If more DATA_DIR-scoped API test files start flaking under shared workers, reset modules before importing singleton db/config modules instead of relying on process env mutation alone
Tested: git diff --check; npm run typecheck; npm run repo:drift-check; npm test
Not-tested: Live OAuth/provider flows outside the mocked test harness
The root checkout already contained a large in-progress proxy, transformer,
and settings refactor. This commit captures that exact work on a dedicated
local dev branch without disturbing the original dirty worktree, so the
changes can be continued or split from dev instead of living only as
uncommitted state.
Constraint: The source checkout was already dirty on another branch, so the move had to preserve that worktree while excluding local .idea and .omx residue from the commit
Rejected: Switch the dirty root checkout directly to dev | risked disrupting the original branch and in-progress state
Confidence: medium
Scope-risk: broad
Reversibility: clean
Directive: Treat this as a transfer snapshot; fix the failing oauth/accountTokens/normalized/docs test regressions before merge or publication
Tested: git diff --check; npm run typecheck; npm run repo:drift-check
Not-tested: Clean test pass; npm test currently fails in oauth routes, accountTokens sync, shared normalized helpers, gemini route expectations, architecture-boundaries, and docs vitepress config
* Make payload request overrides configurable from Settings
The payload-rules engine already existed in the proxy path, but operators had
no supported runtime UI to manage CPA-style overrides. This wires the setting
through the runtime settings API, adds strict validation, and ships a
low-friction visual editor with an advanced JSON fallback so common cases like
Codex high reasoning can be configured without hand-editing backend state.
Constraint: Must reuse the existing payload-rules execution path and settings table without schema changes
Constraint: Must fit the existing metapi Settings-page patterns rather than introducing a standalone config surface
Rejected: JSON-only editor as the primary UX | too long and too easy to misconfigure for common cases
Rejected: A brand-new backend rule format for the UI | would duplicate the existing payload-rules source of truth
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep visual-rule serialization and advanced JSON editing in sync before extending the rule model
Tested: npm run typecheck
Tested: npx vitest run --root . src/server/services/payloadRules.test.ts src/server/routes/api/settings.events.test.ts src/web/pages/settings.payload-rules.test.tsx
Tested: npm run repo:drift-check
Not-tested: Manual browser interaction smoke on the Settings page in this clean PR worktree
* Prevent payload-rules saves from violating review expectations
Code review surfaced two operator-facing edge cases in the first payload-rules
PR: successful rule parsing could still be committed before a later validation
error, and visual edits could silently overwrite unsaved advanced JSON. This
stages payload-rules until the request finishes validating and adds an explicit
confirmation gate before visual actions discard unsynced advanced edits.
Constraint: Keep the fix narrow to the reviewed payload-rules feature without refactoring unrelated runtime-settings atomicity behavior
Rejected: Silently skipping draft resync while keeping visual edits active | would leave save semantics ambiguous when advanced JSON stays dirty
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: If the settings endpoint ever moves toward full request atomicity, re-check this staged payload-rules write path against the broader strategy
Tested: npm run typecheck
Tested: npx vitest run --root . src/server/routes/api/settings.events.test.ts src/web/pages/settings.payload-rules.test.tsx
Tested: npm run repo:drift-check
Not-tested: Manual browser smoke of the confirmation dialog text
The usage aggregation projector was written with SQLite/Postgres conflict builders, but Render can switch the runtime DB to MySQL from saved settings. That made the startup scheduler throw before the service stayed healthy.
This change branches the projector's checkpoint and aggregate writes to MySQL-safe onDuplicateKeyUpdate calls while preserving the existing SQLite/Postgres onConflict behavior. A mocked MySQL regression test now omits the onConflict builders entirely so future regressions fail fast.
Constraint: Drizzle conflict APIs are dialect-specific and the projector starts during boot
Rejected: Disable the projector on MySQL | would keep admin analytics stale and hide the real incompatibility
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: New usage-aggregation writes must branch on runtimeDbDialect before using conflict helpers
Tested: npx vitest run --root . src/server/services/usageAggregationService.test.ts src/server/services/usageAggregationService.mysql.test.ts
Tested: npm run typecheck:server
Tested: npm run repo:drift-check
Tested: npm run build:server
Not-tested: Live MySQL projector pass against a deployed runtime database
* Recognize more LobeHub-backed model brands so marketplace filters stop collapsing into other
Metapi was leaving many models under the uncategorized bucket even though the
current LobeHub icon set already contains matching provider or vendor brands.
This change moves brand detection into one shared registry, expands the rules
with the user-reported gaps that have real upstream icon coverage, and makes
server-side brand blocking reuse the same classification result as the web UI.
Constraint: Existing workspace has unrelated in-progress edits, so this commit stages only the brand-classification slice
Constraint: Brand matching must stay consistent between marketplace display and server-side blocked-brand filtering
Rejected: Add more ad hoc rules separately in web and server files | would keep the drift bug alive
Rejected: Blindly import every LobeHub slug as a model brand | would pollute filters with tool and app brands
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep new brand additions in src/server/shared/modelBrand.ts and let web/server consume that single source of truth
Tested: npm test -- src/web/components/BrandIcon.test.ts src/server/services/brandMatcher.test.ts src/web/pages/Models.marketplace-text.test.tsx src/web/components/BrandIcon.architecture.test.ts src/web/components/BrandIcon.render.test.tsx
Tested: npm run typecheck
Tested: npm run repo:drift-check
Not-tested: Live browser verification against the user's current marketplace dataset after deployment
* Preserve provider brand blocking without changing display-brand priority
The first pass unified brand detection but server-side blocked-brand checks
started relying on the single display brand. Mixed model paths such as
`deepinfra/meta-llama/...` and `azureai/gpt-4o` still need provider markers to
match global blocked brands even while the UI continues to show the preferred
vendor label.
Constraint: Marketplace display should keep the current preferred vendor-first brand label
Constraint: Server-side blocked-brand matching must still recognize provider and vendor markers on the same model id
Rejected: Reorder display-brand priority to make provider blocks pass | would regress the marketplace labels the PR was improving
Rejected: Reintroduce a separate server-only matcher | would recreate the web/server drift this PR removed
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Use getMatchingBrandNames for blocking-style checks and reserve getBrand for single-label display surfaces
Tested: npm test -- src/server/shared/modelBrand.test.ts src/server/services/brandMatcher.test.ts src/web/components/BrandIcon.test.ts src/web/pages/Models.marketplace-text.test.tsx src/web/components/BrandIcon.architecture.test.ts src/web/components/BrandIcon.render.test.tsx
Tested: npm run typecheck
Tested: npm run repo:drift-check
Not-tested: Live GitHub review-thread auto-resolution behavior after push
* Honor legacy blocked brand casing when canonicalizing global brand filters
The shared brand matcher now supports multi-match provider blocking, but the
blocked-brand rule parser still required exact display-case brand names. That
meant imported or legacy settings like `openai` or `aws bedrock` were silently
ignored even though they should map to known brands.
Constraint: Existing saved settings may contain lower-case or whitespace-variant brand names from imports or older data
Constraint: The server should keep emitting canonical display names after normalizing user-provided blocked brand entries
Rejected: Assume the settings UI always writes canonical brand strings | does not cover imported or hand-edited configuration
Rejected: Normalize only at the settings route | would still leave hydrated legacy config values ineffective in modelService
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep blocked-brand parsing tolerant of case and spacing drift, but always return canonical brand names for downstream use
Tested: npm test -- src/server/services/brandMatcher.test.ts src/server/shared/modelBrand.test.ts src/web/components/BrandIcon.test.ts src/web/pages/Models.marketplace-text.test.tsx src/web/components/BrandIcon.architecture.test.ts src/web/components/BrandIcon.render.test.tsx
Tested: npm run typecheck
Tested: npm run repo:drift-check
Not-tested: Manual settings UI round-trip with legacy lowercase globalBlockedBrands payloads
* Make admin reads snapshot-first so heavy pages stay responsive
This introduces a lightweight snapshot/cache layer for the busiest
admin reads, splits dashboard and proxy-log loading into smaller
independent surfaces, and moves accounts/site/dashboard snapshots
toward service-owned read paths.
The UI now loads dashboard summary, dashboard insights, site stats,
accounts, and proxy-log metadata as separate progressive reads while
keeping existing feature coverage intact.
Constraint: Must keep the full admin feature set and core metrics intact
Constraint: Interactive admin flows should trend toward <=3s without relying on OpenResty-only band-aids
Rejected: Delete heavy pages or reduce statistics depth | user explicitly requires unchanged functionality
Rejected: Keep one giant proxy-log/dashboard response | preserves current timeout-prone read shape
Confidence: medium
Scope-risk: moderate
Directive: Preserve old endpoints during the migration and keep new snapshot/query-v2 surfaces as the preferred frontend path
Tested: npm run typecheck
Tested: npx vitest run --root . src/server/routes/api/accounts.snapshot-v2.test.ts src/server/routes/api/stats.snapshot-v2.test.ts src/server/routes/api/stats.proxy-logs.test.ts src/server/routes/api/stats.siteStatus.test.ts src/server/routes/api/accounts.todayRewardFallback.test.ts
Tested: npm run build
Tested: npm run repo:drift-check
Not-tested: Live cita/cita1 latency measurements against the production dataset
* Keep snapshot services lean after the admin read-path rollout
The snapshot-first rollout now has fresh production evidence on cita1, so this follow-up trims two no-op patterns without changing behavior: a single-element Promise.all and a catch-then-rethrow refresh chain. Keeping the service layer lean reduces noise before final Ralph sign-off and redeploy verification.
Constraint: Keep deployed behavior unchanged while satisfying the mandatory deslop pass
Rejected: Wider refactor across route files | would broaden the verified surface right before production comparison closeout
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep snapshot services focused on cache/read orchestration; avoid reintroducing decorative async wrappers
Tested: npm run typecheck; npm run build; targeted vitest snapshot/proxy-log/account suites; npm run repo:drift-check
Not-tested: Public Cloudflare path, which still appears to hit a different origin than cita1
* Keep admin snapshots warm without introducing Redis
This moves the new admin read path beyond process-local caching by persisting
snapshot payloads in a dedicated admin_snapshots table and warming them in
the background. It also migrates remaining lightweight internal callers away
from the heaviest legacy reads and marks legacy endpoints as deprecated
instead of deleting them abruptly.
Constraint: User explicitly rejected adding Redis for admin snapshot acceleration
Rejected: Redis-backed shared cache | adds a forbidden runtime dependency
Rejected: Delete legacy endpoints immediately | still breaks tests, docs, and compatibility callers
Confidence: high
Scope-risk: moderate
Directive: Remove legacy admin endpoints only after internal callers, docs, and compatibility windows are cleared
Tested: npm run typecheck
Tested: npm run build
Tested: npx vitest run --root . src/server/services/adminSnapshotStore.test.ts src/server/routes/api/accounts.snapshot-v2.test.ts src/server/routes/api/stats.snapshot-v2.test.ts src/server/services/startupInfo.test.ts
Tested: npm run repo:drift-check
Not-tested: cita1 staging redeploy on this follow-up commit
* Collapse admin reads onto the canonical API paths
This removes the parallel v2 read surface and routes the snapshot-first
implementation back through the original admin API names. The frontend now
uses the canonical paths again while the backend keeps the newer snapshot,
persistence, and warm-refresh machinery behind those routes. Along the way it
also fixes the main review findings around cache coalescing, active-site scope,
atomic snapshot writes, and auth-header handling.
Constraint: User requested no Redis and no long-lived duplicate admin read endpoints
Rejected: Keep both legacy and v2 endpoints indefinitely | doubles maintenance and confuses the canonical contract
Rejected: Delete old names without remapping callers | would create avoidable frontend and tooling churn
Confidence: high
Scope-risk: moderate
Directive: Treat /api/stats/dashboard, /api/accounts, /api/stats/proxy-logs, and the site stats routes as the canonical admin read contract unless a future breaking-change plan says otherwise
Tested: npm run typecheck
Tested: npm run build
Tested: npx vitest run --root . src/server/services/adminSnapshotStore.test.ts src/server/routes/api/accounts.snapshot-v2.test.ts src/server/routes/api/stats.snapshot-v2.test.ts src/server/routes/api/stats.proxy-logs.test.ts src/server/services/startupInfo.test.ts
Tested: npm run repo:drift-check
Not-tested: cita1 staging redeploy on this canonical-route follow-up commit
* Keep canonical admin analytics fast without raw log scans
This lands the usage-aggregate path as the single implementation behind the canonical admin routes, adds the aggregate schema/migration artifacts, and reconnects dashboard/site stats warming to the new service so the PR no longer sits in a half-projector state.
It also updates the cache and aggregate tests to reflect the real runtime behavior and validates the current branch on cita1 staging, which showed the app-side snapshot path is fast while the remaining 1.5s staging slowness lives on the NodePort path rather than inside the aggregate queries themselves.
Constraint: Must preserve the canonical old-path admin API contract while moving reads onto aggregate-backed snapshots
Constraint: Must not add Redis or a parallel second analytics implementation
Rejected: Keep usageAggregationProjectorService as a separate path | would preserve duplicate ownership and keep type wiring split
Rejected: Judge staging performance only from NodePort timings | hid that pod-local snapshot reads were already fast
Confidence: medium
Scope-risk: moderate
Directive: If dashboard or insights feels slow again on staging, compare NodePort and pod-local timings before changing aggregate queries
Tested: npm run typecheck; npx vitest run --root . src/server/services/adminSnapshotStore.test.ts src/server/services/usageAggregationService.test.ts src/server/services/snapshotCacheService.test.ts src/server/routes/api/accounts.snapshot-v2.test.ts src/server/routes/api/stats.snapshot-v2.test.ts src/server/routes/api/stats.proxy-logs.test.ts src/server/services/startupInfo.test.ts; npm run build; npm run repo:drift-check; cita1 staging deploy plus canonical-route curl benchmarks
Not-tested: cita1 real ingress/OpenResty path; multi-replica projector concurrency
* Prevent admin aggregate drift under concurrent warmers
This follow-up hardens the snapshot-first admin analytics path by adding a persisted projector lease, exact 24-hour dashboard counting, and regression coverage for lease contention and time-window behavior.
The goal is to keep the canonical admin routes correct when multiple workers are alive and to avoid quietly overstating proxy24h by reading hour buckets as if they were an exact sliding window.
Constraint: Projector coordination must work through the existing checkpoint table without adding Redis or a second background system
Constraint: Schema artifacts must stay additive under the repo's generated bootstrap/upgrade pipeline
Rejected: Remove the left-prefix day/hour indexes in this slice | schema:contract rejects non-additive index removals, so that cleanup needs a broader migration strategy
Confidence: high
Scope-risk: moderate
Directive: Keep projector claims DB-backed; do not regress to process-local inFlight-only dedupe for aggregate writes
Tested: npm run typecheck; npx vitest run --root . src/server/services/usageAggregationService.test.ts src/server/routes/api/stats.snapshot-v2.test.ts; npm run build; npm run repo:drift-check
Not-tested: Multi-host live contention beyond synthetic lease test; resolving CodeRabbit index-nit without additive-schema support
* Keep CI aligned with canonical admin snapshots
The admin snapshot cutover left a mixed test surface: MySQL schema artifacts still
classified admin_snapshots.snapshot_key as JSON, canonical /api/accounts callers
were asserted as arrays, and a few proxy tests mocked routeRefreshWorkflow too
narrowly, which leaked missing exports into later OAuth runs.
This commit narrows the schema metadata heuristic, updates the checked-in schema
artifacts and their regression tests, adds lightweight web test compat helpers
for canonical snapshot APIs, and preserves actual routeRefreshWorkflow exports so
full-suite workers no longer poison unrelated tests. OAuth/account token suites
also reset DATA_DIR-backed modules safely to avoid stale singleton config.
Constraint: checked-in schema artifacts drive cross-dialect CI parity checks
Constraint: vitest module mocks can leak across files in the same worker when exports are omitted
Rejected: rerun schema:generate only | additive guard rejects the logical-type correction for snapshot_key
Rejected: keep per-test ad hoc snapshot mocks | repeated breakage across pages needed one compat helper
Confidence: high
Scope-risk: moderate
Directive: when a test uses vi.resetModules with mutable config or DATA_DIR state, re-import those modules after the reset instead of mutating a stale top-level singleton
Tested: npm test
Tested: npm run typecheck
Tested: npm run build
Tested: npm run repo:drift-check
Tested: npm run test:schema:parity
Not-tested: live MySQL schema parity against a local MySQL instance
* Keep CodeQL from re-flagging already-limited routes
GitHub Advanced Security still marked three admin routes as missing rate limits
because the repository uses a custom Fastify preHandler wrapper instead of one
of the middleware packages modeled by the CodeQL query. The routes are already
throttled in production, so this change adds narrow in-source suppressions on the
three affected handlers to record that the check is a false positive instead of a
real missing guard.
Constraint: repo policy forbids adding new dependencies just to satisfy a static analyzer
Constraint: the custom createRateLimitGuard preHandler is runtime-correct but not modeled by CodeQL's missing-rate-limiting query
Rejected: add fastify-rate-limit | would introduce a new dependency and duplicate existing guard behavior
Rejected: dismiss alerts only in GitHub UI | leaves the false-positive rationale detached from the source line that keeps re-triggering the alert
Confidence: medium
Scope-risk: narrow
Directive: if these routes are ever converted away from createRateLimitGuard, remove or revisit these suppressions rather than assuming they remain safe forever
Tested: npm run typecheck
Tested: npm run repo:drift-check
Not-tested: full npm test rerun after comment-only suppression change
* Use legacy CodeQL suppression syntax for custom rate guards
The modern source suppression comment did not affect GitHub Advanced Security's
missing-rate-limiting check on these Fastify handlers. Switch the three known
false-positive locations to the legacy LGTM suppression syntax, keeping the
custom preHandler throttling in place while matching the syntax that CodeQL
still honors for in-source suppressions.
Constraint: createRateLimitGuard remains the runtime source of truth for throttling in this repo
Constraint: GitHub's missing-rate-limiting query does not model the repo's custom Fastify preHandler wrapper
Rejected: dismiss the alerts in the GitHub UI | the rationale should stay attached to the source line that keeps re-triggering the alert
Rejected: add a recognized rate-limit dependency | repo rules disallow a new dependency for analyzer appeasement alone
Confidence: medium
Scope-risk: narrow
Directive: if GitHub eventually honors the newer codeql[] syntax here, remove whichever suppression form becomes redundant so the file keeps one clear rationale
Tested: Not run locally (comment-only suppression syntax change)
Not-tested: GitHub CodeQL rerun after switching to LGTM suppression syntax
* Keep admin routes clean after CodeQL false-positive triage
The repository's custom createRateLimitGuard preHandlers are the actual runtime
throttles for these routes. Since GitHub Advanced Security still did not honor
the attempted in-source suppressions, remove the temporary comments and keep the
source focused on runtime behavior while the false-positive alerts are tracked
operationally in GitHub.
Constraint: the custom Fastify preHandler remains the only real throttling mechanism in this repo
Rejected: leave non-functional suppression comments in source | they imply enforcement that GitHub did not actually honor
Confidence: high
Scope-risk: narrow
Directive: if CodeQL adds custom Fastify preHandler modeling later, prefer removing any alert dismissals over reintroducing comment noise here
Tested: Not run locally (comment-only cleanup)
Not-tested: GitHub checks after alert dismissal
Metapi was leaving many models under the uncategorized bucket even though the
current LobeHub icon set already contains matching provider or vendor brands.
This change moves brand detection into one shared registry, expands the rules
with the user-reported gaps that have real upstream icon coverage, and makes
server-side brand blocking reuse the same classification result as the web UI.
Constraint: Existing workspace has unrelated in-progress edits, so this commit stages only the brand-classification slice
Constraint: Brand matching must stay consistent between marketplace display and server-side blocked-brand filtering
Rejected: Add more ad hoc rules separately in web and server files | would keep the drift bug alive
Rejected: Blindly import every LobeHub slug as a model brand | would pollute filters with tool and app brands
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep new brand additions in src/server/shared/modelBrand.ts and let web/server consume that single source of truth
Tested: npm test -- src/web/components/BrandIcon.test.ts src/server/services/brandMatcher.test.ts src/web/pages/Models.marketplace-text.test.tsx src/web/components/BrandIcon.architecture.test.ts src/web/components/BrandIcon.render.test.tsx
Tested: npm run typecheck
Tested: npm run repo:drift-check
Not-tested: Live browser verification against the user's current marketplace dataset after deployment
* normalize primary site url handling
* fix site create redirect test selector
* fix sub2api gemini model discovery and site detect race
* fix sub2api antigravity fallback and test selectors
* fix site modal wording test
* guard blur detect against manual platform changes
cita-777