fix(sync-gbrain): seed DATABASE_URL from ~/.gbrain/config.json into gbrain spawns

[thehashrocket]

Summary

/sync-gbrain fails inside any project whose .env.local defines its own DATABASE_URL (Next.js, Prisma, Rails, etc.). gbrain auto-loads the project's .env.local from cwd via dotenv, picks up the app's local-Postgres URL, and tries to authenticate there with the password from ~/.gbrain/config.json. Auth fails. Code and memory stages crash; only the git-push brain-sync stage survives.

Repro

cd <Next.js project with DATABASE_URL=postgresql://postgres:***@localhost:5433/app in .env.local>
/sync-gbrain
# → ERR  code         source registration failed: gbrain not configured (run /setup-gbrain)
# → ERR  memory       gbrain import exited 1: password authentication failed for user "postgres"
# → OK   brain-sync   curated artifacts pushed

gbrain doctor run from the same cwd reports URL from env:DATABASE_URL — confirming gbrain is picking up the project's value, not the configured one. gbrain doctor from /tmp connects cleanly.

Fix

New helper buildGbrainEnv() in bin/gstack-gbrain-sync.ts:

• Reads ~/.gbrain/config.json.
• Returns { ...process.env, DATABASE_URL: cfg.database_url }.
• Skipped when GSTACK_RESPECT_ENV_DATABASE_URL=1 is set (escape hatch for users whose brain genuinely lives in the project DB).
• One-line log when it overrides a caller value.

Threaded through every gbrain spawn in runCodeImport, runMemoryIngest, and runBrainSyncPush. lib/gbrain-sources.ts already exposed an env option for exactly this case — wired up ensureSourceRegistered(... { env }) and sourcePageCount(id, env).

Why not just mutate process.env?

Tried that first. Doesn't work in Bun: child_process.spawnSync children receive Bun's startup env, not runtime mutations. Verified with a minimal repro — printenv in a Bun-spawned child returns the original .env.local value even after process.env.DATABASE_URL = .... Must pass env: explicitly. There's already a comment in lib/gbrain-sources.ts noting this exact caveat.

Verification

End-to-end run from the repro project after the patch:

[gbrain-sync] seeded DATABASE_URL from ~/.gbrain/config.json (overrode value from caller env / .env.local)
[gbrain-sync] mode=incremental engine=supabase

gstack-gbrain-sync (incremental):
  OK    code         registered + synced gstack-code-org-fd8a2521-d5b972 (page_count=575) (289.7s)
  OK    memory       gbrain import: 9 imported, 0 unchanged, 0 failed (36.2s)
  OK    brain-sync   curated artifacts pushed (0.4s)

  3 ok, 0 error, 0 skipped

Upstream gbrain consideration

The underlying root cause is on gbrain's side: the CLI shouldn't let CWD .env.local override its own ~/.gbrain/config.json. Either config wins over CWD dotenv, or gbrain namespaces its connection var (GBRAIN_DATABASE_URL) so collisions are impossible. This patch is a gstack-side workaround; happy to also file a gbrain bug.

Test plan

• bun run bin/gstack-gbrain-sync.ts --help parses
• bun run bin/gstack-gbrain-sync.ts --dry-run emits the seed-env log line
• Real incremental sync from inside a Next.js project: all 3 stages OK (575 pages indexed)
• Escape hatch: GSTACK_RESPECT_ENV_DATABASE_URL=1 bun run ... returns process.env unchanged (verified by reading the buildGbrainEnv code path)

🤖 Generated with Claude Code

[garrytan]

Thank you @thehashrocket. Cherry-picked into v1.40.0.0 with attribution preserved (Co-Authored-By: Jason Shultz). Restructured per codex review into a centralized lib/gbrain-exec.ts helper; widened scope to also thread env into the gstack-memory-ingest grandchild. See commit 0fb7fa6. Consolidated v1.40.0.0 fix wave lands as PR #1547. Closing this one as superseded — your contribution is recognized in CHANGELOG.md and the commit metadata.