5.1.1

Combines the 5.0.0 functional jump with a chromium-scale tokensave_dead_code perf fix.

What's new (since 4.14.11)

9 new MCP tools

• tokensave_read — mode-aware file read (full / lines / map / signatures) with cross-session cache. A re-call on an unchanged file returns a ~30-token {"unchanged": true, …} stub.
• tokensave_outline — flat list of every top-level symbol in a file, with optional kind filter.
• tokensave_implementations — every type implementing a trait, or every body of a method.
• tokensave_unsafe_patterns — .unwrap() / .expect() / panic! / todo! / unimplemented! / unsafe { } sites with an in_test flag.
• tokensave_diagnostics — runs cargo / tsc / pyright and returns structured errors mapped to graph nodes. Forces .tokensave/target/ so it can't race with interactive cargo runs.
• tokensave_config — TOML / JSON queries by dotted key path. DB-free.
• tokensave_signature_search — find functions / methods by return type, params, async flag.
• tokensave_constructors — every struct-literal site with missing_fields relative to the current definition.
• tokensave_field_sites — partition every .<field> reference into reads and writes.

Console-friendly tokensave bench

Default tokensave bench is now a fixed-width colored table — compact k/M units, tier-colored savings (green ≥80 %, yellow ≥50 %, red <50 %). --json unchanged.

Schema v9

• read_cache table — backs tokensave_read's cross-session cache.
• Contains edges denormalized into nodes.parent_id — single forward-only migration. Cleaner queries; extractors keep emitting Contains and the storage layer hoists at insert time. Recovery path for downstream consumers still reading Contains directly: tokensave sync -f.

Performance — chromium-scale tokensave_dead_code

find_dead_code rewritten as a three-step resolve via two TEMP tables (temp.test_markers, temp.test_annotated_targets). Probed against the real chromium DB (7.5 GB, 4.4 M nodes, 411 K annotates edges):

call	before	after
dead_code {}	timeout	2.45 s
dead_code {limit: 10}	timeout	1.10 s
dead_code {path: "src"}	timeout	1.10 s

Pre-fix was 5/5 TIMEOUT at the 25 s ceiling and cascade-poisoned every subsequent MCP tool. On scirs (76 K annotation_usage): 0.6 s vs the 0.097 s pre-4.14.8 baseline — regression-acceptable given chromium went from >25 s to <1.1 s steady-state. Inline comment on find_dead_code documents three prior failed shapes (pre-4.14.8 wildcard, 4.14.8 CTE inline, single-temp-table) so future refactors don't repeat them.

Upgrade detection no longer races CI

fetch_latest_version() now verifies the current platform's archive is uploaded to the release before announcing it. Releases whose binaries haven't finished building are silently skipped — clients see "you're up to date" until the asset matching their platform is present. The existing per-attempt download check (which already produced a clear error) is preserved as belt-and-suspenders.

Migration

• Schema v9 is forward-only. First sync after upgrade auto-applies the migration, populates parent_id from existing Contains rows, and deletes those rows.
• External SQLite consumers reading the edges table should switch from kind='contains' filters to nodes.parent_id joins.
• If anything else goes sideways: tokensave sync -f rebuilds the graph from source under the new schema.