@alignfirst/openclaw-test 0.21.0 → 0.22.0

package/README.md

@@ -45,8 +45,8 @@ OPENROUTER_API_KEY=sk-or-… # Required for an OpenRouter agent or judge.
 OPENCLAW_WORKSPACE_DIR=/path/to/your/openclaw-workspace
 
 # Model catalog: full LiteLLM refs. `run --model` picks by bare id (suffix after the last "/").
-OPENCLAW_TEST_MODELS=anthropic/claude-sonnet-4-6,custom-openrouter/qwen/qwen3.6-plus
-OPENCLAW_DEFAULT_TEST_MODEL=claude-sonnet-4-6
+OPENCLAW_TEST_MODELS=anthropic/claude-sonnet-5,custom-openrouter/qwen/qwen3.8-flash
+OPENCLAW_DEFAULT_TEST_MODEL=claude-sonnet-5
 
 ```
 
@@ -84,8 +84,8 @@ npm run env:up                                                     # (optional)
 npm run e2e -- --channel all <scenario>                             # one scenario, both channels
 npm run e2e -- --channel all --all                                  # every scenario, both channels
 npm run e2e -- --channel discord-mock <scenario>                    # restrict to one channel
-npm run e2e -- --channel all --model qwen3.6-plus <scenario>        # pick a model by bare id
-npm run e2e -- --channel all --model claude-sonnet-4-6,qwen3.6-plus <s>  # a comma list of bare ids
+npm run e2e -- --channel all --model qwen3.8-flash <scenario>        # pick a model by bare id
+npm run e2e -- --channel all --model claude-sonnet-5,qwen3.8-flash <s>  # a comma list of bare ids
 npm run e2e -- --channel all --model all <scenario>                 # run every model in OPENCLAW_TEST_MODELS
 npm run e2e -- --channel all --iterations 5 <scenario>              # repeat each (scenario, channel) pair 5×
 npm run e2e -- --channel all --iterations 5 --max-failures 1 <s>    # abort a pair after >1 failure
@@ -121,10 +121,10 @@ Assert on `conversation.id` / `threadId`, not envelope formatting.
 
 ## Judge model
 
-Defaults to `anthropic/claude-haiku-4-5`. Override it via `OPENCLAW_TEST_JUDGE_MODEL` on
-the `runner` service (set in your consumer overlay). Direct Anthropic refs use
-`anthropic/<model>`; OpenRouter refs use `openrouter/<model>`, for example
-`openrouter/anthropic/claude-haiku-4.5`. The judge is **not** an OpenClaw agent — don't
+Defaults to `openrouter/anthropic/claude-haiku-4.5`, billed to `OPENROUTER_API_KEY`. Override it
+via `OPENCLAW_TEST_JUDGE_MODEL` on the `runner` service (set in your consumer overlay). OpenRouter
+refs use `openrouter/<model>`; direct Anthropic refs use `anthropic/<model>`, for example
+`anthropic/claude-haiku-4-5`. The judge is **not** an OpenClaw agent — don't
 configure it in `openclaw.json`.
 
 ## Attribution

package/dist/env-cli.js

@@ -20,7 +20,7 @@ const RUN_USAGE = `usage: openclaw-test run --channel <id|id,id,…|all> [<scena
 
   Scenario selection is required: either a positional list or --all (mutually exclusive).
   --model <id|id,id,…|all>
-                      select the agent model(s): a bare id (e.g. claude-sonnet-4-6), a
+                      select the agent model(s): a bare id (e.g. claude-sonnet-5), a
                       comma list of bare ids, or "all". Defaults to OPENCLAW_DEFAULT_TEST_MODEL.
                       The catalog is OPENCLAW_TEST_MODELS (.env.local), a comma list of full
                       provider/model refs; the bare id is the suffix after the last "/".
@@ -407,6 +407,49 @@ function setupHostEnv(packageDir) {
         process.env.ASSISTANT_GID = String(process.getgid?.() ?? 1000);
     absolutizePathVarsFromEnvFile(projectDir);
     applyPathDefaults(projectDir);
+    pinBuilderReadingLocalImages();
+}
+// The consumer Dockerfile resolves the base image with `FROM paleo/openclaw-test-base:<tag>`,
+// a tag that exists only in the local image store. Only a `docker`-driver builder reads that
+// store: a `docker-container` or `remote` builder tries the registry instead and fails with an
+// opaque `pull access denied`. `docker build` always picks the docker driver, so the base image
+// itself is unaffected, but `docker compose build` follows the selected builder — so a developer
+// who selected a container builder for everything else cannot build the consumer image.
+// Point BUILDX_BUILDER at the docker-driver builder, whose name follows the Docker context and
+// is only `default` on the default one. An explicit BUILDX_BUILDER wins, as does the selected
+// builder when it already reads the store.
+function pinBuilderReadingLocalImages() {
+    if (process.env.BUILDX_BUILDER)
+        return;
+    const builders = listBuilders();
+    if (builders.some((builder) => builder.current && builder.driver === "docker"))
+        return;
+    const local = builders.find((builder) => builder.driver === "docker");
+    if (local)
+        process.env.BUILDX_BUILDER = local.name;
+}
+// `buildx ls --format json` emits one builder per line. Tolerate every failure: an older Docker
+// without the flag, or a malformed line, leaves the selection untouched and the build reports
+// its own error.
+function listBuilders() {
+    const r = spawnSync("docker", ["buildx", "ls", "--format", "json"], { encoding: "utf8" });
+    if (r.status !== 0 || !r.stdout)
+        return [];
+    return r.stdout
+        .split("\n")
+        .map((line) => line.trim())
+        .filter(Boolean)
+        .flatMap((line) => {
+        try {
+            const parsed = JSON.parse(line);
+            if (!parsed.Name || !parsed.Driver)
+                return [];
+            return [{ name: parsed.Name, driver: parsed.Driver, current: parsed.Current === true }];
+        }
+        catch {
+            return [];
+        }
+    });
 }
 // Defaults relative to the consumer's project dir, applied after `.env.local` so
 // explicit values win. Doing this in the CLI (not via `${VAR:-default}` in

package/dist/judge.js

@@ -1,7 +1,7 @@
 import Anthropic from "@anthropic-ai/sdk";
 import OpenAI from "openai";
 import { extractTaggedBlock } from "./parse-tagged-json.js";
-const DEFAULT_JUDGE_MODEL = "anthropic/claude-haiku-4-5";
+const DEFAULT_JUDGE_MODEL = "openrouter/anthropic/claude-haiku-4.5";
 const DEFAULT_MAX_TOKENS = 1024;
 const RESULT_TAG = "result-json";
 let cached;

package/dist/transcript-store.js

@@ -1,4 +1,5 @@
 import { DatabaseSync } from "node:sqlite";
+import { zstdDecompressSync } from "node:zlib";
 export function readConversationSessions(dbPath, sinceMs, conversationId, threadIds) {
     let db;
     try {
@@ -51,19 +52,28 @@ function readNodeMessages(db, node, sinceMs) {
 }
 function readSessionMessages(db, sessionId, sinceMs) {
     const rows = db
-        .prepare("SELECT event_json FROM transcript_events WHERE session_id = ? AND created_at >= ? ORDER BY seq")
+        .prepare("SELECT event_json, event_zstd FROM transcript_events" +
+        " WHERE session_id = ? AND created_at >= ? ORDER BY seq")
         .all(sessionId, sinceMs);
     const messages = [];
     for (const row of rows) {
         let event;
         try {
-            event = JSON.parse(row.event_json);
+            event = JSON.parse(decodeTranscriptEvent(row));
         }
         catch {
             continue;
         }
-        if (event.type === "message" && event.message !== undefined)
+        if (event?.type === "message" && event.message !== undefined)
             messages.push(event.message);
     }
     return messages;
 }
+/** Since OpenClaw 2026.9.6, an event of 1 KiB or more is stored zstd-compressed, `event_json` NULL. */
+function decodeTranscriptEvent(row) {
+    if (row.event_json !== null)
+        return row.event_json;
+    if (row.event_zstd === null)
+        throw new Error("transcript event without a payload");
+    return zstdDecompressSync(row.event_zstd).toString("utf8");
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@alignfirst/openclaw-test",
-  "version": "0.21.0",
+  "version": "0.22.0",
   "description": "Dockerised regression-test framework for OpenClaw workspaces: bus, scenario driver, judge, Compose stack.",
   "keywords": [
     "openclaw",
@@ -44,10 +44,10 @@
     "openclaw": "*"
   },
   "dependencies": {
-    "@alignfirst/openclaw-channel-mock-core": "0.10.0",
-    "@alignfirst/openclaw-discord-mock": "0.6.0",
-    "@alignfirst/openclaw-slack-mock": "0.6.0",
-    "@anthropic-ai/sdk": "~0.122.0",
+    "@alignfirst/openclaw-channel-mock-core": "0.10.2",
+    "@alignfirst/openclaw-discord-mock": "0.6.2",
+    "@alignfirst/openclaw-slack-mock": "0.6.2",
+    "@anthropic-ai/sdk": "~0.127.0",
     "openai": "~7.8.0"
   },
   "devDependencies": {
@@ -55,6 +55,6 @@
     "openclaw": "~2026.9.3",
     "rimraf": "~6.1.3",
     "typescript": "~7.0.2",
-    "vitest": "~4.1.11"
+    "vitest": "~5.0.1"
   }
 }

package/templates/.env.local.example

@@ -7,8 +7,8 @@ OPENROUTER_API_KEY=
 # place the provider/ prefix appears. `run --model <id>` and OPENCLAW_DEFAULT_TEST_MODEL
 # use the bare id (the suffix after the last "/"). `--model all` runs every entry and
 # needs the API key for each referenced provider.
-OPENCLAW_TEST_MODELS=anthropic/claude-sonnet-4-6,custom-openrouter/qwen/qwen3.6-plus
-OPENCLAW_DEFAULT_TEST_MODEL=claude-sonnet-4-6
+OPENCLAW_TEST_MODELS=anthropic/claude-sonnet-5,custom-openrouter/qwen/qwen3.8-flash
+OPENCLAW_DEFAULT_TEST_MODEL=claude-sonnet-5
 
 # Required: host path to the OpenClaw workspace (mounted into the gateway).
 OPENCLAW_WORKSPACE_DIR=