gbrain upgrade runs gbrain apply-migrations automatically. Most users
won't need to do anything else. If you hit issues or want to verify the
upgrade succeeded, run the steps below.
v0.34.0.0 ships the recursive code-intelligence layer on top of v0.33.3's MCP-exposed code-callers/callees/def/refs foundation:
Recursive walks (NEW for agents):
code_blast(symbol, depth=5, max_nodes=200)— recursive callers walk with depth grouping. Use BEFORE editing any function.code_flow(entry_point, depth=8, max_nodes=200)— recursive callees walk with terminal-sink tagging. Use when tracing how a request flows.code_traversal_cache_clear— admin op for invalidating cached walks.
Foundation extensions:
- W1: receiver-type resolution at extraction time.
obj.method()now resolves toClass::methodfor the 3 MUST patterns in JS/TS/TSX + Python (import, this/self, new C()). Bare-token emit on miss (pre-W1 behavior). - W2: new
importsandreferencesedge types. JS/TS/TSX + Python get imports; TS gets type-position references. Ruby/Go/Rust/Java stay at calls-only. - W3b:
code_traversal_cachetable for memoizing blast/flow results. Cache invalidation honors the D3 cluster-generation counter. - W7:
src/core/eval-capture-graph.ts— pure-function metrics module for replay-driven regression checks.
gbrain upgrade runs:
bun install(refreshes deps; no new external deps in v0.34)gbrain post-upgrade→gbrain apply-migrations→ migration v56 (code_traversal_cache_v0_34) creates the cache table on Postgres and PGLite.
If your agent uses code symbol-resolution heavily, also run:
gbrain edges-backfill --all-sourcesThis walks every content_chunks row whose edges_backfilled_at is
NULL or older than EDGE_EXTRACTOR_VERSION_TS (bumped by v0.34 to
re-walk every chunk with the new receiver-resolution + imports +
references emission shape). Resumable; Ctrl-C is safe.
gbrain doctor --json | jq '.checks.code_intel_coverage'
gbrain call code_blast --symbol performSync
gbrain call code_flow --entry_point runCyclePer the plan's explicit slip-handling clause, the following ship in v0.34.1 after operator brain validation:
- W4–5 Leiden clusters. Cluster ship gate (≤0.03 clusters/node) requires real brain data; v0.34.0.0 ships without clusters.
gbrain wiki— depends on clusters.gbrain blast/gbrain flowCLI thin-shims. Usegbrain call code_blast --symbol <name>for v0.34.0.0; CLI wrappers come with v0.34.1.- eval-capture wiring — the metric module ships in v0.34.0.0; capture/replay dispatch on tool comes with v0.34.1.
File an issue at https://github.com/garrytan/gbrain/issues with:
- output of
gbrain doctor - contents of
~/.gbrain/upgrade-errors.jsonlif it exists - which step broke