sync --strategy code dropped on first sync via performFullSync

[rayers]

Summary

gbrain sync --strategy code --source <id> silently does a markdown-only import on first sync (no anchor commit yet). The --strategy flag is parsed correctly but dropped when performFullSync calls runImport, which is hardcoded to walk markdown only via collectMarkdownFiles. As a result, registering a fresh code source and running first sync produces 0 code pages — no error, just silent under-coverage. code-def / code-refs / code-callers / code-callees then return 0 hits for every symbol in the repo.

Steps to reproduce

# Start with a fresh code source
gbrain sources add gstack-code-myrepo --path /path/to/myrepo --federated

# First sync with --strategy code
cd /path/to/myrepo
gbrain sync --source gstack-code-myrepo --strategy code --no-embed

# Output: 0 pages imported, N skipped (where N = the number of .md files
#         that already exist as duplicates in default source)
# Expected: thousands of code pages from .ts/.py/.java/.c/etc.

Verify:

gbrain code-def <known_symbol>  # returns count: 0

Root cause

src/commands/sync.ts:847 — performFullSync calls runImport from ./import.ts:

const { runImport } = await import('./import.ts');
const importArgs = [repoPath];
if (opts.noEmbed) importArgs.push('--no-embed');
if (fullConcurrency > 1) importArgs.push('--workers', String(fullConcurrency));
const result = await runImport(engine, importArgs, { commit: headCommit });

Note that opts.strategy is in scope but never threaded into importArgs. runImport then uses collectMarkdownFiles(repoPath) which only enumerates *.md. The walkSyncableFiles(strategy) path that incremental mode uses (sync.ts:111-134) is bypassed entirely on first sync.

The dry-run branch above (sync.ts:858) has the same drop — collectMarkdownFiles regardless of opts.strategy.

Affected behavior

• First-time gbrain sources add <id> --path <repo> followed by gbrain sync --strategy code produces no code pages.
• Subsequent invocations of sync --strategy code after the first still hit the same path because last_commit was never advanced.
• Workaround gbrain reindex-code --source <id> --yes prints No code pages to reindex (chicken-and-egg — needs code pages first).
• setting strategy=code directly in the source's config row in postgres doesn't help — the code path of performFullSync ignores the source config too because it uses runImport.

Suggested fix

Thread opts.strategy through to runImport (or call a strategy-aware walker directly inside performFullSync instead of delegating to the markdown-only runImport). The incremental path (walkSyncableFiles(repoPath, cb, strategy)) already does the right thing; ideally performFullSync would use the same walker.

Same fix needs to apply to the dry-run branch.

Environment

• gbrain v0.30.1 (commit dffb607)
• Engine: postgres 16.13 (pgvector + pg_trgm)
• macOS 26.3.1 (build 25D771280a), bun 1.3.10
• Use case: Dividia NVR repo (~7000 files: Java + Python + C + TypeScript + bash). Goal was enabling mcp__gbrain__code_def for cross-codebase symbol lookup.

Workarounds tried

1. gbrain sources add ... --strategy code — no --strategy flag on sources add.
2. Setting config.strategy = 'code' directly in postgres — ignored by runImport.
3. gbrain sync --source <id> --strategy code --full --no-pull — same bug.
4. gbrain reindex-code --yes --source <id> — needs pre-existing code pages.

Currently no clean workaround short of writing a custom TS script that imports importCodeFile from gbrain internals and walks the repo manually.