Bugfix/upload - #222
Closed
romanbsd wants to merge 50 commits into
Closed
Bugfix/upload#222romanbsd wants to merge 50 commits into
romanbsd wants to merge 50 commits into
Conversation
release: promote main
chore: release release
release: promote main
release: promote main
release: promote main
release: promote main
release: promote main
release: promote main
release: promote main
release: v1.8.1
release: v1.8.2
release: v1.9.0
release: v1.10.0
release: v1.11.0
release: v1.12.0
release: v1.13.0
release: v1.14.0
release: v1.15.0
release: v1.15.1
release: v1.15.2
release: v1.15.3
Implement XMPP gateway and expose agents' capabilities
* feat: add general contractor project workflow * refactor: DRY contact name and primary contact derivation * fix(api): satisfy anti-slop lint in deal contact and gc field updates * fix: address PR review feedback on project workflow * fix: require customers for construction projects * docs: map database to construction workflow * refactor: remove unsupported construction fields * fix: align customer deletion copy * docs: track construction changes * refactor: restore upstream application UI * docs: update construction change map * fix: preserve upstream project mappings * test: isolate auth environment defaults * feat: retain minimal construction CRM changes * docs: align construction mapping terms * fix: restore CRM vocabulary in construction POC * fix: consolidate additive CRM migration --------- Co-authored-by: Roman Shterenzon <roman.shterenzon@gmail.com>
Integrate Kaneo's project management into the CRM without reimplementation and without a second schema. - Kaneo is a git submodule at vendor/kaneo pointing at the crm-integration branch of the romanbsd/kaneo fork, which carries the CRM deltas. - One Postgres schema, owned by Prisma. The 32 kaneo tables are generated from an ORM-agnostic abstract model (packages/kaneo-domain), with a Drizzle binding and a parity test proving the generated schema matches what kaneo's code expects. kaneo tables use snake_case physical columns; shared auth tables keep the CRM's camelCase. activity and invitation are renamed to task_activity and workspace_invitation to avoid collisions. - kaneo's own Hono controllers and web UI are mounted and served: dev:kaneo boots the API against the shared database, serves the web SPA, proxies /api and bridges /ws. kaneo's startup Drizzle migrations are gated. - One session cookie (crm.session_token) valid at both apps, one shared Better Auth identity, and CRM roles mapped onto kaneo's on sign-in. - The eve agent reads projects and tasks through Prisma and drives writes through kaneo's extracted controller functions directly, acting as the workspace owner. Documented in vendor/FORK-DELTA.md, adrs/kaneo.md and docs/kaneo-integration.md.
A POST /push-tokens endpoint stores a device FCM token for the signed-in user (upsert by token, transferred to the latest owner), and DELETE /push-tokens removes it if it belongs to the caller. Tokens are validated with zod at the boundary (platform is ios|android). Add the push_token table, the module wiring, and e2e coverage for the unauthenticated paths.
* fix: preserve app build environment * fix(auth): allow mixed Google sign-in domains * feat(app): add public legal pages
* feat: add customer and project asset storage * fix: validate system message before upload replay * docs: set JobSteward pull request repository * refactor: keep asset code files within 200 lines
Raise the declared toolchain from bun 1.3.12 to 1.4.2 in packageManager and devEngines. The lockfile format is unchanged; bun 1.4.2 reads and writes it identically and a frozen install resolves cleanly.
chore: support bun 1.4.x
* chore(deps): upgrade all dependencies to latest versions Upgrades eve, ai, next, react, typescript, prisma, better-auth, biome, oxlint, turbo, and the remaining manifests to their latest versions. Fixes incompatibilities introduced by the upgrades: - eve 0.52.3: ClientSessions.attach() replaces session(), ChannelFrom/ChannelSource replace SendFn, respond() is a separate channel method. - @pierre/diffs 1.4.1: new Editor/FileOptions generics. - typescript 7 (tsgo): explicit node types in agent-xmpp tsconfigs. - better-auth 1.7.3: regenerated the oauth-provider patch. - prisma 7 capped at 7.10.0 (8.x requires a config-file migration). * refactor: dedupe builder dispatch parsing and agent model fallback Extract shared builder input-response predicate and delivery-part builder so custom-agent-dispatch parses the submission JSON once instead of up to three times. Add defaultAgentModelResult() to replace five duplicated fallback object literals across agent entrypoints. Fix indentation and non-null assertions flagged by biome in packages/agent-xmpp/core/src/schema.ts. Add @better-auth/core to packages/auth dependencies. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * fix: satisfy anti-slop lint on defaultAgentModelResult Drop the explicit return type annotation so TypeScript keeps the literal-typed inference from DEFAULT_AGENT_MODEL instead of widening it to string/number. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
* feat: add native multi-tenancy * fix: tenant-scope customer assets * update ci.yml * fix: declare @better-auth/core as direct dependency in @crm/auth slack-grant.ts imports @better-auth/core/context but the package was never listed as a dependency, relying on phantom hoisting. Multiple @better-auth/core versions in the tree made resolution nondeterministic and CI hoisted a version without the symlink, failing check-types with TS2307. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * fix: sort test imports to satisfy biome lint Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * fix: sort imports and format per biome across rebase diff Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * remove workflows * fix lint * fix: satisfy anti-slop lint boundary-parsing rules Fixes the lint:slop CI failure. Replaces inline Record<string, unknown> and typeof-based narrowing with named types and instanceof/Zod parsing at real I/O boundaries. Moves api-key-principal and adds active-organization-claim parsers into packages/validation, matching the repo's parse-at-boundary convention. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013nRJf8FVnwdjmSwgBt7tuX * fix: give real-worker schema test more headroom on CI The "starts source workers with the declared loader" test spins up a real worker_threads worker (tsx-loaded), unlike the mocked worker in schema-worker-lineage.test.ts. Its default 500ms budget is tuned for production safety, not CI cold-start latency, so a loaded CI runner timed it out at 502ms. Passes an explicit 5s timeout for this one integration-style test only; production default is unchanged. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013nRJf8FVnwdjmSwgBt7tuX * fix: raise both bun's per-test timeout and the worker budget Previous fix only bumped validateJsonBounded's internal timeoutMs to 5000ms, missing that bun:test's own default per-test timeout is also 5000ms. Both fired at once, so bun killed the test before the internal promise could even reject on its own terms. Raise the test timeout to 20s and the internal budget to 15s, giving real headroom on a resource-constrained CI runner. Verified against a Linux bun 1.4.2 container. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013nRJf8FVnwdjmSwgBt7tuX * fix: stop spawning schema workers with a Node tsx loader under bun Root cause of the hanging schema.test.ts: createSchemaWorker passed execArgv: ["--import", "tsx"] when spawning the source-mode worker. Everything in this repo runs under bun, not Node, and bun already runs .ts worker files natively without any loader hook. Verified with a raw Worker spawn that bun runs the .ts file fine with no execArgv at all. On Linux x64 (the GitHub Actions runner) that unnecessary flag combination silently hung the worker forever - neither 5s nor 15s budgets helped because the worker never came online, not because it was slow. Confirmed against a Linux bun 1.4.2 container before and after. Removes the now-dead tsx devDependency and the execArgv option, and reverts the test back to the default timeout now that the real fix lands. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013nRJf8FVnwdjmSwgBt7tuX * diag: surface the real worker failure reason in CI logs Two prior fixes (longer timeouts, removing execArgv) did not resolve this test's failure on the GitHub Actions Linux x64 runner, and it is not reproducible on macOS ARM64 or a native Linux ARM64 bun container (x64 emulation via QEMU crashes bun itself with a memory-exhaustion assertion, unrelated to the real bug, so it gives no signal). Every prior CI failure only showed "promise rejected" with no reason. Logs worker error/exit events to stderr in schema.ts (previously silent) and logs the actual rejection in the test before failing, so the next CI run tells us what is actually happening instead of us guessing again. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013nRJf8FVnwdjmSwgBt7tuX * diag: trace schema worker startup step by step Prior diagnostic showed no error/exit event ever fires for the real schema.test.ts worker on CI - it just silently times out, meaning the worker either never starts, or starts and hangs somewhere before registering the message handler, with a code-0 exit event we were not logging. Switches schema-worker.ts to dynamic imports with a log after each step (isMainThread/parentPort, ajv import, protocol import, Ajv construction, handler registration, message receipt), adds uncaughtException/unhandledRejection handlers, and logs the worker's "online" event and every exit code (not just nonzero) on the parent side. Confirmed locally the full sequence logs and the test passes. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013nRJf8FVnwdjmSwgBt7tuX * fix: stop leaking a mock.module Worker across test files Root cause, confirmed with a minimal local repro: bun 1.4.2's mock.restore() does not actually undo mock.module("node:worker_threads", ...) for other test files sharing the same bun test process. Once schema-worker-lineage.test.ts mocked node:worker_threads to return a FakeWorker, schema.test.ts's plain "./schema.js" import of the real Worker silently resolved to that FakeWorker instead - a class that never emits online/message/error, so every validateJsonBounded() call just sat there until our own internal timeout fired. That is why every prior fix (longer timeouts, dropping the tsx execArgv) failed identically: the worker was never real, so no amount of waiting or loader changes could help. It only reproduced on the GitHub Actions runner because import/module-cache ordering between the two test files happened to differ there; a standalone repro (two files, one mocking node:worker_threads and restoring, the other freshly importing it after) reproduces the same leak locally. Fixes this at the root instead of routing around it: schema.ts now constructs its worker through an overridable factory (setWorkerFactory), so schema-worker-lineage.test.ts can inject its FakeWorker with a plain reassignment scoped to its own "./schema.ts?worker-lineage" module instance, never touching bun's global module registry. Removes mock.module/mock.restore entirely from that test. Reverts the temporary diagnostic logging from the last two commits now that the cause is known. Verified 5x locally and 3x in a Linux bun 1.4.2 container. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013nRJf8FVnwdjmSwgBt7tuX --------- Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
No file changes — master already contains equivalent content from a prior sync. This merge commit only records the shared ancestry so future syncs diff cleanly. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
feat: organize R2 asset keys by organization and project
* feat: add project appointments and asset management * fix: refresh activity caches after appointment changes * test: use Prisma connection for appointment lock races
* feat: replace public upload workflows with asset resources * fix: keep database context out of browser bundles
|
@romanbsd is attempting to deploy a commit to the Comp AI - PoC Team on Vercel. A member of the Team first needs to authorize it. |
Contributor
|
Retargeted this onto
Nothing is wrong with your branch. If the diff now shows commits that are already on git fetch origin main
git rebase origin/main
git push --force-with-lease |
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
3 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.
Summary by cubic
Fixes asset uploads by replacing the public upload workflows with organization- and project-scoped asset resources served at root resource paths. The branch also merges the platform's multi-tenant, Kaneo, and XMPP gateway work into
release, so the diff is far larger than the upload fix alone.Migration
vendor/kaneois now a git submodule; rungit submodule update --init.master; the auto-PR, PR-base, PR-title, and release workflows were removed, and the Postgres credentials changed frompostgres/postgrestocrm/crm.Written for commit fdeb4fa. Summary will update on new commits.