@ataraxy-labs/sem 0.23.0 → 0.23.1

package/CHANGELOG.md

@@ -4,6 +4,22 @@ All notable changes to sem are documented in this file.
 
 ## [Unreleased]
 
+## [0.23.1] - 2026-08-22
+
+### Fixed
+
+- **CSV, JSON, and Vue entities no longer collide on generated ids with entities from other files.** Their id-generation scheme is now disambiguated per plugin, closing a gap where two entities could silently collapse onto the same id and one would drop out of the graph.
+- **Entities from non-code files (Markdown, TOML, YAML, JSON, CSV, Vue/Svelte) now carry accurate byte ranges**, so tools that rely on byte offsets (extraction, editing, highlighting) work correctly for these file types instead of getting an inaccurate span.
+- **Markdown headings that appear inside a fenced code block are no longer parsed as real document headings.**
+- **TypeScript/JavaScript entity byte spans now include a leading `export` keyword when present**, so extracting an exported declaration's exact source text no longer drops the `export ` prefix. Facts schema v4 — existing caches rebuild automatically on first use.
+- **`sem entities` no longer opens the git repository through libgit2 on every call.** That was a fixed per-call cost regardless of file size, disproportionately noticeable on small-file lookups; it's now only paid when actually needed.
+- **sem-mcp's `query` and `limit` parameters on the `entities` and `context` tools now work correctly** (previously ignored).
+
+### Added
+
+- **sem-mcp: new `find` and `grep` tools**, giving MCP clients the same fast entity-lookup and trigram-accelerated search already available from the CLI (`sem find`, `sem grep`).
+- **sem-mcp: `entities` and `context` tools accept `format=json`**, returning structured JSON instead of human-readable text for callers that want to parse results programmatically.
+
 ## [0.23.0] - 2026-08-22
 
 ### Changed
@@ -55,7 +71,7 @@ All notable changes to sem are documented in this file.
 
 ### Performance
 
-- **Cold graph builds are 11-30% faster and peak RSS is down 17-40% on giant corpora, versus 0.21.** Measured end-to-end on the shipped release binary: home-assistant-core 5.9s, TypeScript monster 11.2s, dotnet-runtime 46.7s, llvm-project 34.7s, linux 35.2s cold full-CLI (warm rebuilds: 0.2-1.3s). Every corpus improved on both the engine-only and full-CLI metrics; full methodology and per-corpus numbers are in `RESOLUTION-PROFILE.md`'s FINALE section.
+- **Cold graph builds are 11-30% faster and peak RSS is down 17-40% on giant corpora, versus 0.21.** Measured end-to-end on the shipped release binary: home-assistant-core 5.9s, TypeScript monster 11.2s, dotnet-runtime 46.7s, llvm-project 34.7s, linux 35.2s cold full-CLI (warm rebuilds: 0.2-1.3s). Every corpus improved on both the engine-only and full-CLI metrics.
 - **C#/C++ builds now skip re-parsing files whose facts are already known**, closing the last gap in precomputed-facts reuse (JS/TS/Python/Go/Java/Rust already had it). A per-file gate proves a corpus-wide invariant — no entity's parent lives in a different file — before trusting precomputed facts wholesale, so this needed no facts-schema change. Measured on dotnet-runtime: reparse time drops from 10.6s to 65ms.
 - **Parsed file facts now persist to disk as a content-addressed corpus**, so a build that has seen a file's exact content before warm-starts it instead of re-parsing from scratch, even in a fresh process. Fixed a regression where checking a shared corpus against a large number of prior contributors got slower as the corpus grew (one repo's known-content rebuild was measured 332% slower against a 7.9GB shared corpus than a 556MB one); it now costs the same regardless of corpus size.
 - **`sem context` regained a fast tier it had lost, by reading each entity's body from its own file at an indexed byte span instead of walking and hydrating the whole corpus.** A prior cascade of cache removals deleted the old fast path along with a correctness bug it had, but left `sem context` always doing a full corpus load — measured on the TypeScript monster 1.11s down to 48ms, on a mid-size repo 0.16-0.28s down to 4.6ms. Verified byte-identical against the always-correct full-load path across both entity- and file-scoped lookups; still declines (never approximates) on a stale cache, an ambiguous name, or any entity missing a span.

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@ataraxy-labs/sem",
   "mcpName": "io.github.Ataraxy-Labs/sem",
-  "version": "0.23.0",
+  "version": "0.23.1",
   "description": "npm wrapper for the sem CLI. Downloads the matching release binary and exposes the sem command in node_modules/.bin.",
   "license": "MIT OR Apache-2.0",
   "type": "module",