feat(postgres): multi-project isolation for the shared embedded-PG cluster (task project_id, Approach A)

[TrinaryCompute]

Rebase feature/postgres onto v0.57 + fix embedded-PG task→merge blockers + multi-project isolation

Builds on the embedded-PostgreSQL migration (#1793). Rebases it onto v0.57.0, fixes the blockers that stopped an autonomous task → merge cycle from completing in embedded-PG mode, and closes the multi-project isolation gap. Builds clean (core/engine/dashboard, 0 tsc errors); validated live on a 6-project instance.

Rebase

main was 154 commits ahead of feature/postgres; 43 conflicts resolved (kept the async-Drizzle PG rewrite, ported main's behavior changes into it).

Blocker fixes (one family: cwd vs rootDir)

• recoverStaleTransitionPendingImpl threw SQLite … not available in backend mode → added the missing backendMode branch.
• Specified tasks stuck "unplanned" → startup-factory.ts dropped options.rootDir (store rooted at process.cwd()); now if (projectId && !rootDir).
• Merge never landed → drainMergeQueue() bound git to process.cwd(); now store.getRootDir() ?? config.workingDirectory.

Multi-project isolation

project.tasks had no project_id (SQLite isolated per-file; consolidating to shared PG tables dropped that partition). Added project_id + index, threaded the filter through the claim query / scheduler / merger / dedup / list API / id sequence.

Per-project config isolation + live concurrency hot-reload

• project.config was a single shared row (CHECK id=1) → shared taskPrefix/caps + a cross-project mega-context on every triage. Added project_id PK + configScope(). Validated: distinct per-project prefixes; triage input 243K → 41–91K tokens.
• Global concurrency limit hot-reloads (ProjectManager subscribes to concurrency:changed) — PUT /api/global-concurrency applies live, no restart.

Happy to split any of these out if the maintainers prefer smaller PRs.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 99b4430a-17e3-4c60-bbf6-b82a2b7812c0

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Greptile Summary

This PR adds multi-project isolation for the shared embedded PostgreSQL setup. The main changes are:

• Adds project_id partitioning for live tasks and project config rows.
• Threads project scoping through task reads, search, merge leases, and creation paths.
• Updates startup migration behavior for legacy SQLite data.
• Adds live global concurrency reload handling in the engine.

Confidence Score: 4/5

This is close, but the archive isolation gap should be fixed before merging.

• Live task and config paths are mostly scoped by project now.
• Archived task listings can still return rows from other projects.
• The remaining issue needs a schema and query update for the archive storage path.

packages/core/src/postgres/schema/project.ts and the archive table/helpers

Security Review

Archived task rows can still be visible across projects because the canonical archive table remains shared without project ownership.

Important Files Changed

Filename	Overview
packages/core/src/postgres/schema/project.ts	Adds live task, config, and legacy archived-task partition columns, but the canonical archive storage path remains unpartitioned.
packages/core/src/postgres/startup-factory.ts	Binds backend task stores to project IDs and updates legacy migration stamping for project-scoped rows.
packages/core/src/task-store/async-persistence.ts	Stamps and filters live task rows with the bound project ID.
packages/core/src/task-store/async-settings.ts	Moves bound project config operations to the new project_id key.
packages/core/src/task-store/async-search.ts	Adds project scoping to full-text and fallback task search.
packages/core/src/task-store/async-merge-coordination.ts	Scopes merge lease acquisition through the owning task's project.

_{Reviews (3): Last reviewed commit: "feat(postgres): multi-project isolation ..." | Re-trigger Greptile}

[gsxdsm]

Archive isolation P1 — fixed on feature/postgres (commit 46fea60)

Confirmed: the archived board read the shared archive.archived_tasks cold-storage table unfiltered (the PR's project.archived_tasks.project_id only covers the ID allocator's table). Since this PR merged while the fix was in flight, it landed directly on the base branch:

• project_id added to archive.archived_tasks (schema + baseline DDL + index), stamped by both archive writers from the bound layer's projectId
• listArchivedTaskEntriesPage / getArchivedRowCount / listArchivedTasks / filterArchived / searchArchivedTasks (+ the tsvector/LIKE search helpers) take an optional projectId and filter to it; the archived-board impl threads layer.projectId; id-keyed get/delete stay unscoped (ids are globally unique via the distributed allocator)
• the drift self-heal now supports schema-qualified entries, so existing embedded-PG databases gain the column on boot
• gate regression test: two projects' archive entries, page/count/membership/search each scoped (store-list.pg.test.ts)

Gates green: PG 23 files / 99 tests, engine-core 294, core typecheck clean.