fix(update): recover from OpenClaw size-drop rejection

CHANGELOG.md

@@ -34,6 +34,7 @@ All notable changes to ClawRouter.
   - Endpoint count 84 → 83, removed the Chat section, added the required-param pre-check note up front.
   - Fixed param names that the agent would otherwise feed incorrectly and trip the pre-check 400: `social/mindshare` is `q` + `interval` (not `project` + `window`), `search/*` family is `q` (not `query`), `token/holders` + `token/transfers` need `address` AND `chain`, `onchain/gas-price` needs `chain`, `onchain/tx` needs `hash` + `chain`, exchange family universally needs `pair`, market family uses `symbol`, prediction-market endpoints use their specific identifier params (`market_slug`, `event_slug`, `condition_id`, `market_ticker`, `event_ticker`, `ticker`, `address`), and `fund/ranking` + `project/defi/*` need `metric`.
   - Reworded the example flows that previously used wrong param names (`search/project?query=ethena` → `?q=ethena`, mindshare example fixed).
+
 - **Phone skill — voice/call `from` is now optional with server-side auto-pick.** blockrun moved `from` from required → optional on `/v1/voice/call`. After payment verification the server picks a caller-ID from the wallet's owned numbers: 0 active → `403 no_active_number` with buy-first hint, exactly 1 → auto-used, 2+ → `400 ambiguous_from` listing all candidates. The prior skill said "Otherwise Bland picks one" — wrong, and would have made agents leave `from` off and confuse users with the 403/400 responses. Updated `skills/phone/SKILL.md` with the actual auto-pick rule table and the ownership-mismatch 403 behavior.
 - **No code change for `real_face_asset_id`, no new partner tools, no proxy whitelist change.** Surf + Phone are both base+skill integrations (per the v0.12.193 rule). All blockrun upstream changes here flow through the existing `proxyPaidApiRequest` path transparently.
 
@@ -87,7 +88,7 @@ All notable changes to ClawRouter.
   - **Bug 1 — phantom $0.54 charge on 4xx voice POST.** First smoke test POSTed `/v1/voice/call` with empty `{}` body to exercise routing without spending money. BlockRun returned 400 (Zod validation: "expected string, received undefined"). The wallet wasn't charged, but the telemetry hook saw `paymentStore.amountUsd = 0` and fell back to `estimatePhoneCost("/v1/voice/call") = $0.54`. Stats would record a phantom voice call. Fix: gate the fallback on `upstream.status` being 2xx — any 4xx/5xx skips the fallback and logs `$0`.
   - **Bug 2 — GET poll miscounted as another $0.54 voice call.** After placing a real call, polling `GET /v1/voice/call/{call_id}` for transcript status (free upstream) was being logged at $0.54 because the longest-prefix match on `voice/call/` triggered the same fallback row as the initiating POST. Every 30s poll would inflate stats by $0.54. Fix: also gate the fallback on `req.method === "POST"` — GET polls log `$0`.
   - **Refactor**: gate logic was originally inline inside `proxyPaidApiRequest`. Pulled it out into `resolvePhoneTelemetryCost(args)` so the rules are independently testable (the call site is now four lines passing an args bag through the helper). Adds 8 vitest cases covering: paid-amount-wins, 4xx phantom guard, GET poll guard, 5xx guard, missing-method guard, non-phone-passthrough, and the original "successful POST with empty paymentStore → fallback" path. Without the helper extraction, locking these gates in tests would have required a full integration test with a mocked upstream — too heavy for telemetry-only logic.
-- **Tests** — new `src/proxy.phone-routing.test.ts` (regex matching for /v1/phone/*, /v1/voice/*, /v1/voice/call/{id} poll, plus negative case for /v1/phonebook), `src/proxy.phone-pricing.test.ts` (longest-prefix matching + the 8 `resolvePhoneTelemetryCost` gate cases above), `src/parse-call-args.test.ts` (both flag forms, quoted task spans, E.164 first-token detection). Total 31 new test cases; all 569 vitest tests green; typecheck + lint clean.
+- **Tests** — new `src/proxy.phone-routing.test.ts` (regex matching for /v1/phone/\*, /v1/voice/\*, /v1/voice/call/{id} poll, plus negative case for /v1/phonebook), `src/proxy.phone-pricing.test.ts` (longest-prefix matching + the 8 `resolvePhoneTelemetryCost` gate cases above), `src/parse-call-args.test.ts` (both flag forms, quoted task spans, E.164 first-token detection). Total 31 new test cases; all 569 vitest tests green; typecheck + lint clean.
 - **Smoke test record** (free-tier verification before the real call): list-numbers ($0.001) returned an existing wallet-owned number `+15707043521` (PA, expires 2026-06-15); lookup ($0.01) on that same number returned full Twilio carrier metadata (`type: nonFixedVoip`, `carrier_name: Twilio - SMS/MMS-SVR`); negative test `/v1/phonebook/test` correctly rejected by the partner regex (502 from chat-completion fallback rather than partner routing); CLI table formatting + expiry-warning logic verified by `clawrouter phone numbers list`.
 
 ---

README.md

@@ -307,16 +307,16 @@ Verify phone numbers and place AI-powered outbound voice calls directly from cha
 
 Calls are **fire-and-forget**: the request returns a `call_id` and `poll_url` immediately. The call itself runs in the cloud for up to 30 minutes. Poll `GET /v1/voice/call/{call_id}` (or `clawrouter share`/transcripts dashboard) to retrieve the transcript and recording when status is `completed`.
 
-| Operation                         | Provider | Price                        |
-| --------------------------------- | -------- | ---------------------------- |
-| Phone lookup (carrier, line type) | Twilio   | $0.01                        |
-| Fraud check (SIM-swap, fwd)       | Twilio   | $0.05                        |
-| Buy phone number (30-day lease)   | Twilio   | $5.00                        |
-| Renew lease (+30 days)            | Twilio   | $5.00                        |
-| List wallet's owned numbers       | Twilio   | $0.001                       |
-| Release a number                  | Twilio   | free                         |
-| **AI voice call (≤30 min)**       | Bland.ai | **$0.54 flat per call**      |
-| Poll call status / transcript     | Bland.ai | free                         |
+| Operation                         | Provider | Price                   |
+| --------------------------------- | -------- | ----------------------- |
+| Phone lookup (carrier, line type) | Twilio   | $0.01                   |
+| Fraud check (SIM-swap, fwd)       | Twilio   | $0.05                   |
+| Buy phone number (30-day lease)   | Twilio   | $5.00                   |
+| Renew lease (+30 days)            | Twilio   | $5.00                   |
+| List wallet's owned numbers       | Twilio   | $0.001                  |
+| Release a number                  | Twilio   | free                    |
+| **AI voice call (≤30 min)**       | Bland.ai | **$0.54 flat per call** |
+| Poll call status / transcript     | Bland.ai | free                    |
 
 **CLI for wallet-owned numbers:**
 
@@ -354,11 +354,11 @@ Surf is BlockRun's unified crypto data API — **84 endpoints across 13 domains*
 
 ClawRouter ships Surf as a **skill, not as typed wrappers**. The proxy whitelists `/v1/surf/*` so any call through the local proxy is paid x402 from the same wallet; the agent reads `skills/surf/SKILL.md` for the endpoint catalog and crafts the HTTP call. No `blockrun_surf_*` tool definitions to maintain; a new Surf endpoint requires zero ClawRouter release.
 
-| Tier | Cost      | Examples                                                      |
-| ---- | --------: | ------------------------------------------------------------- |
-| 1    | **$0.001**| prices, rankings, lists, news                                 |
-| 2    | **$0.005**| orderbooks, candles, search, wallet details, social mindshare |
-| 3    | **$0.020**| on-chain SQL / query / schema, chat completions               |
+| Tier |       Cost | Examples                                                      |
+| ---- | ---------: | ------------------------------------------------------------- |
+| 1    | **$0.001** | prices, rankings, lists, news                                 |
+| 2    | **$0.005** | orderbooks, candles, search, wallet details, social mindshare |
+| 3    | **$0.020** | on-chain SQL / query / schema, chat completions               |
 
 **Usage (HTTP):**
 

scripts/update.sh

@@ -324,32 +324,164 @@ for stale in "$HOME/.openclaw/extensions/clawrouter.backup."* "$HOME/.openclaw/e
   [ -d "$stale" ] && rm -rf "$stale"
 done
 
+apply_scoped_model_trim() {
+  local rejected_path="$1"
+  if [ -z "$rejected_path" ] || [ ! -f "$rejected_path" ] || [ ! -f "$CONFIG_PATH" ]; then
+    return 1
+  fi
+
+  CONFIG_PATH="$CONFIG_PATH" REJECTED_CONFIG_PATH="$rejected_path" node <<'NODE'
+const fs = require('fs');
+
+const activePath = process.env.CONFIG_PATH;
+const rejectedPath = process.env.REJECTED_CONFIG_PATH;
+
+function fail(message) {
+  console.log(`  Skipped scoped config trim: ${message}`);
+  process.exit(1);
+}
+
+function byteSize(value) {
+  return Buffer.byteLength(JSON.stringify(value ?? null, null, 2));
+}
+
+function objectKeys(value) {
+  return value && typeof value === 'object' && !Array.isArray(value) ? Object.keys(value).sort() : [];
+}
+
+function clone(value) {
+  return JSON.parse(JSON.stringify(value));
+}
+
+function getBlockrunModels(config) {
+  return config?.models?.providers?.blockrun?.models;
+}
+
+const active = JSON.parse(fs.readFileSync(activePath, 'utf8'));
+const rejected = JSON.parse(fs.readFileSync(rejectedPath, 'utf8'));
+
+const activeTopKeys = objectKeys(active);
+const rejectedTopKeys = objectKeys(rejected);
+if (activeTopKeys.join('\0') !== rejectedTopKeys.join('\0')) {
+  fail('top-level config keys changed');
+}
+
+for (const key of ['auth', 'channels', 'gateway', 'plugins', 'models']) {
+  if (!(key in active) || !(key in rejected)) fail(`missing required ${key} section`);
+}
+
+const activeModels = getBlockrunModels(active);
+const rejectedModels = getBlockrunModels(rejected);
+if (!Array.isArray(activeModels) || !Array.isArray(rejectedModels)) {
+  fail('blockrun model list is missing or invalid');
+}
+
+if (activeModels.length <= rejectedModels.length) {
+  fail(`model count did not shrink (${activeModels.length} -> ${rejectedModels.length})`);
+}
+
+if (rejectedModels.length < 20 || rejectedModels.length > 100) {
+  fail(`unexpected curated model count (${rejectedModels.length})`);
+}
+
+for (const key of activeTopKeys) {
+  if (key === 'models') continue;
+  const delta = Math.abs(byteSize(active[key]) - byteSize(rejected[key]));
+  if (delta > 2048) fail(`non-model section changed too much: ${key}`);
+}
+
+const activeWithoutModelList = clone(active.models);
+const rejectedWithoutModelList = clone(rejected.models);
+activeWithoutModelList.providers.blockrun.models = [];
+rejectedWithoutModelList.providers.blockrun.models = [];
+const residualModelDelta = Math.abs(byteSize(activeWithoutModelList) - byteSize(rejectedWithoutModelList));
+if (residualModelDelta > 4096) {
+  fail('models section changed beyond the blockrun model list');
+}
+
+const totalDrop = byteSize(active) - byteSize(rejected);
+const modelListDrop = byteSize(activeModels) - byteSize(rejectedModels);
+if (totalDrop <= 0 || modelListDrop / totalDrop < 0.65) {
+  fail('size drop is not primarily from the blockrun model list');
+}
+
+active.models.providers.blockrun.models = rejectedModels;
+const tmpPath = `${activePath}.tmp.${process.pid}`;
+fs.writeFileSync(tmpPath, JSON.stringify(active, null, 2));
+fs.renameSync(tmpPath, activePath);
+
+console.log(
+  `  ✓ Applied scoped BlockRun model-list trim (${activeModels.length} -> ${rejectedModels.length})`,
+);
+NODE
+}
+
+handle_openclaw_install_failure() {
+  local exit_code="$1"
+  if [ "$exit_code" -eq 124 ]; then
+    echo "  (install command timed out — this is normal with OpenClaw v2026.4.5)"
+    if [ -f "$PLUGIN_DIR/package.json" ]; then
+      echo "  Plugin package.json is present; treating install as completed before the hang."
+      return 0
+    fi
+    echo "  Plugin package.json is missing after timeout; continuing with direct npm install."
+    OPENCLAW_INSTALL_RECOVERABLE=1
+    return "$exit_code"
+  fi
+
+  if grep -q "Config write rejected: .*size-drop:" "$OPENCLAW_INSTALL_LOG"; then
+    local rejected_path
+    rejected_path=$(node -e "const fs=require('fs'); const text=fs.readFileSync(process.argv[1],'utf8'); const m=text.match(/Rejected payload saved to ([^\\s]+)\\./); if (m) console.log(m[1]);" "$OPENCLAW_INSTALL_LOG")
+    echo "  ⚠ OpenClaw rejected a config size-drop during plugin registration."
+    if apply_scoped_model_trim "$rejected_path"; then
+      echo "  Continuing with direct npm install after the scoped config update."
+    else
+      echo "  Continuing with direct npm install while preserving your existing config."
+    fi
+    OPENCLAW_INSTALL_RECOVERABLE=1
+    return 0
+  fi
+
+  return "$exit_code"
+}
+
 echo "→ Installing latest ClawRouter..."
 # `--force` is required when the plugin is already installed at the same path
 # (which is always true on update). Without it, OpenClaw exits non-zero with
 # "plugin already exists" and our EXIT trap rolls back, stranding the user on
 # the previous version. `--force` is idempotent for both fresh + upgrade flows.
 #
+# OpenClaw can also reject its own config rewrite when it would shrink a large
+# user config. Treat that as recoverable: keep the user's restored config and
+# install ClawRouter files directly from npm below.
+OPENCLAW_INSTALL_RECOVERABLE=0
+OPENCLAW_INSTALL_LOG="$(mktemp)"
+#
 # Run with timeout — openclaw plugins install may hang after printing
 # "Installed plugin: clawrouter" in OpenClaw v2026.4.5 (parallel plugin loading).
 # 120s is enough for slow connections; the install itself completes in ~30s.
 if command -v timeout >/dev/null 2>&1; then
-  timeout 120 openclaw plugins install --force @blockrun/clawrouter || {
+  timeout 120 openclaw plugins install --force @blockrun/clawrouter 2>&1 | tee "$OPENCLAW_INSTALL_LOG" || {
     exit_code=$?
-    if [ $exit_code -eq 124 ]; then
-      echo "  (install command timed out — this is normal with OpenClaw v2026.4.5)"
-      echo "  Plugin was installed successfully before the hang."
-    else
-      exit $exit_code
-    fi
+    handle_openclaw_install_failure "$exit_code" || {
+      [ "$OPENCLAW_INSTALL_RECOVERABLE" = "1" ] || exit $exit_code
+    }
   }
 else
-  openclaw plugins install --force @blockrun/clawrouter
+  openclaw plugins install --force @blockrun/clawrouter 2>&1 | tee "$OPENCLAW_INSTALL_LOG" || {
+    exit_code=$?
+    handle_openclaw_install_failure "$exit_code" || {
+      [ "$OPENCLAW_INSTALL_RECOVERABLE" = "1" ] || exit $exit_code
+    }
+  }
 fi
+rm -f "$OPENCLAW_INSTALL_LOG"
 
 # Install is complete — clear the rollback trap immediately.
 # From this point on, Ctrl+C or errors should NOT roll back the install.
-trap - EXIT INT TERM
+if [ "$OPENCLAW_INSTALL_RECOVERABLE" != "1" ]; then
+  trap - EXIT INT TERM
+fi
 
 # Restore credentials after plugin install (always restore to preserve user's channels)
 if [ -n "$CREDS_BACKUP" ] && [ -d "$CREDS_BACKUP" ]; then
@@ -431,6 +563,18 @@ if [ -d "$PLUGIN_DIR" ] && [ -f "$PLUGIN_DIR/package.json" ]; then
   fi
   INSTALLED_VER=$(node -e "try{const p=require('$PLUGIN_DIR/package.json');console.log(p.version);}catch{console.log('?');}" 2>/dev/null || echo "?")
   echo "  ✓ ClawRouter v${INSTALLED_VER} installed"
+elif [ "$OPENCLAW_INSTALL_RECOVERABLE" = "1" ]; then
+  LATEST_VER=$(npm view @blockrun/clawrouter@latest version 2>/dev/null || echo "")
+  if [ -z "$LATEST_VER" ]; then
+    echo "  ✗ Could not resolve latest ClawRouter version from npm"
+    exit 1
+  fi
+  force_install_from_npm "$LATEST_VER"
+  INSTALLED_VER=$(node -e "try{const p=require('$PLUGIN_DIR/package.json');console.log(p.version);}catch{console.log('?');}" 2>/dev/null || echo "?")
+  echo "  ✓ ClawRouter v${INSTALLED_VER} installed"
+fi
+if [ "$OPENCLAW_INSTALL_RECOVERABLE" = "1" ]; then
+  trap - EXIT INT TERM
 fi
 
 # ── Step 4c: Ensure all dependencies are installed ────────────