chore(release): v0.17.5 (#233)

Author: unohee

CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## 0.17.5 — 2026-07-05
+
+### Added
+
+- **Worker escalation on repeated review feedback** — when the reviewer repeats near-identical revise feedback (the 0.17.4 stagnation signal), the pipeline now escalates the worker once and retries in-session before giving up: `worker.escalateModel` when it differs from what the next iteration would run anyway, plus a reasoning-effort bump to `high` (active with zero config). Same feedback after escalation → early abort as before, with the feedback persisted for the next attempt. Escalation policy (iteration-count + signal) now lives in `workerEscalation.ts`. (INT-2475, #232)
+
 ## 0.17.4 — 2026-07-05
 
 ### Fixed

README.md

@@ -530,11 +530,11 @@ the CLI (fire-and-forget with a short timeout, and failures are silently ignored
 Full version history lives in **[CHANGELOG.md](CHANGELOG.md)** and the
 [GitHub Releases](https://github.com/unohee/OpenSwarm/releases) page.
 
-Latest — **v0.17.4**: the reviewer-revise treadmill is broken — failed
-attempts persist their reviewer feedback and re-attempts start with it
-injected, a repeating reviewer ends the session early, and fan-out winner
-promotion no longer dies on dirty projects. Plus v0.17.3's duplicate-daemon
-prevention and fan-out execution fixes. See CHANGELOG.md for the rest.
+Latest — **v0.17.5**: a repeating reviewer now triggers a one-shot worker
+escalation (higher model and/or `high` effort) before the session gives up —
+on top of v0.17.4's cross-session feedback persistence and fan-out
+dirty-promotion fix, and v0.17.3's duplicate-daemon prevention. See
+CHANGELOG.md for the rest.
 
 ---
 

package-lock.json

@@ -1,6 +1,6 @@
 {
   "name": "@intrect/openswarm",
-  "version": "0.17.4",
+  "version": "0.17.5",
   "description": "Autonomous AI agent orchestrator — Claude, GPT, Codex, and local models (Ollama/LMStudio/llama.cpp)",
   "license": "MIT",
   "type": "module",