prjct-cli 2.44.0 → 2.44.2

package/CHANGELOG.md

@@ -3,13 +3,28 @@
 ## [Unreleased]
 
 ### Fixed
-- **Superseded knowledge can no longer surface as live advice.** `searchFts` (the per-prompt trap cue + the FTS leg of every recall) only filtered soft-deletes — an obsolete decision/gotcha could still inject as current. It now prunes author-declared retirements (`supersedes:` / `superseded-by:` / `duplicates:`) using a mirror-wide dead-id set, catching the case where BM25 never co-returns the superseding entry.
-- **Spanish prompts now produce real retrieval signal.** Keyword extraction was ASCII-only ("búsqueda" → mangled, "qué pasó" → nothing); it now tokenizes unicode, deburrs to match FTS5's diacritic-stripped index, and drops Spanish function words. FTS query sanitizing deburrs too.
+- **The first command after an upgrade no longer stalls.** The post-upgrade re-setup (provider installers, Context7 verification, the per-project migration over the whole projects dir) ran synchronously inside whatever command the user happened to run first — up to ~30s of silent blocking on machines with large project dirs. It now runs in a detached `__post-upgrade` child (the auto-updater pattern): the command returns immediately (measured 31.6s → 0.10s) and a one-line banner says setup is finishing in the background. Explicit `prjct update` keeps its synchronous, progress-printing behavior. Both detached-child internals are now in the daemon shim's skip set — routing them to the daemon silently failed them.
+
+### Internal
+- **Dead-code and anti-pattern sweep** (no user-visible behavior change). Removed: 29 redundant `export default` aliases, the orphaned OAuth token validation/migration family (`tokens.ts` shrinks to the version pin; `system-database.ts` and its write-only MCP-health table die with it), `projectMemory.similar()`, `memoryService.getRecent`/`getByAction`, `hasIndexesAll`, `aggregateHookSignals`, `deriveWorkspaceId`, and five write-only `SkillContext` fields (task/backlog descriptions deliberately never enter the skill body). Deduplicated: the guard label/detail helpers (one canonical pair in `memory/format.ts`), the CLAUDE.md/AGENTS.md routing writers (shared `routing-block.ts` skeleton), and the pathManager test sandbox (one `_setup/path-manager-mock.ts` instead of eight hand-copied patch/restore blocks). Simplified: `selectProvider` (the never-built `userSelected` prompt path removed), static skill frontmatter takes no context by construction.
+
+## [2.44.2] - 2026-06-12
+
+### Bug Fixes
+
+- post-upgrade setup no longer stalls the user's first command (~30s → 0.1s) (#432)
+
+### Maintenance
+
+- dead-code + anti-pattern sweep — DRY/KISS pass (#431)
+
+
+## [2.44.1] - 2026-06-12
+
+### Performance
+
+- token diet — coarse timestamps, quiet state lines, counts-only skill State, instruction dedupe (#430)
 
-### Improved
-- **MCP memory tools retrieve like the CLI.** `prjct_mem_list` used plain recency recall and `prjct_mem_similar` a shared-keyword heuristic — subagents (who do the bulk of the editing) got strictly worse retrieval than `prjct context memory`. Both now run the shared `enrichedRecall` pipeline: FTS-first, semantic blend, usefulness rerank, link expansion, ship attribution.
-- **Digest slots are proven-first.** The session-start and subagent digests picked the 3 most RECENT gotchas/decisions; they now overfetch and rerank by the usefulness ledger, so knowledge that keeps paying off wins the slots.
-- **Push-surfaced knowledge earns ship credit.** The pre-edit guard hook, the prompt trap cue, `prjct guard`, and `prjct_guard` now log surfaced entries against the active task — previously only PULL recalls did, so the most effective gotchas decayed in ranking precisely because the push delivery worked.
 
 ## [2.44.0] - 2026-06-11