Version Packages (#45)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

Author: github-actions[bot]

.changeset/run-all-and-improvements.md

@@ -1,5 +1,18 @@
 # @vercel/agent-eval
 
+## 0.3.0
+
+### Minor Changes
+
+- [#44](https://github.com/vercel-labs/agent-eval/pull/44) [`9f7af62`](https://github.com/vercel-labs/agent-eval/commit/9f7af6276ce0f61c79c31ef66cc47b161c0f0028) Thanks [@gaojude](https://github.com/gaojude)! - Add `run-all` command with fingerprinting, failure classification, and housekeeping.
+
+  - **run-all command**: Auto-discovers `experiments/*.ts` and runs them all with fingerprint reuse, AI failure classification, auto-retry of infra failures, and housekeeping. Now the default when `agent-eval` is invoked with no arguments.
+  - **Content fingerprinting**: Computes SHA-256 fingerprints from eval files + config. Skips evals with matching cached results. Safe to extend model arrays or add new evals.
+  - **Failure classification**: Classifies failed evals as model/infra/timeout using AI via `gateway('anthropic/claude-sonnet-4-5')` with sandboxed tools. Requires `AI_GATEWAY_API_KEY`.
+  - **Housekeeping**: Removes duplicate results, incomplete results, and empty timestamp directories after each experiment.
+  - **--smoke flag**: Picks the first eval alphabetically and runs it once per model for quick setup verification.
+  - **Output naming fix**: Script outputs moved to `outputs/scripts/{name}.txt` to prevent collision with `outputs/eval.txt`.
+
 ## 0.2.0
 
 ### Minor Changes

packages/agent-eval/CHANGELOG.md

@@ -1,6 +1,6 @@
 {
   "name": "@vercel/agent-eval",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "description": "Framework for testing AI coding agents in isolated sandboxes",
   "repository": {
     "type": "git",