feat(performance): add progressive disclosure tests and compositional routing acceptance

.github/workflows/e2e-branch-validation.yaml

@@ -46,13 +46,17 @@ name: E2E / Branch Validation
 #                              OpenAI-compatible endpoint. Creates its own sandbox
 #                              (e2e-msg-compat) on a separate fresh instance.
 #   dashboard-remote-bind    — Verifies opt-in remote dashboard forwards bind 0.0.0.0.
+#   tool-disclosure-performance-smoke — Runs the claim-ineligible hosted-inference
+#                              performance smoke on an ephemeral Brev CPU VM and
+#                              retains its sanitized evidence. (~60 min)
 #   gpu                      — Provisions a Brev GPU VM and runs the Ollama GPU E2E
 #                              sandbox proof suite from source. (~45 min)
 #   all                      — Runs credential-sanitization + telegram-injection, each with
 #                              its own sandbox lifecycle (NOT the independent full journey).
 #
 # Required secrets: BREV_API_KEY + BREV_ORG_ID (or legacy BREV_API_TOKEN), NVIDIA_INFERENCE_API_KEY
-# Instance cost: Brev CPU credits (~$0.10/run for 4x16 instance)
+# Instance cost: Brev CPU credits; most suites use 4x16, while the performance
+# smoke requests 8 vCPU, 32 GB RAM, and 100 GB disk.
 
 on:
   workflow_dispatch:
@@ -73,6 +77,7 @@ on:
           - messaging-providers
           - messaging-compatible-endpoint
           - dashboard-remote-bind
+          - tool-disclosure-performance-smoke
           - gpu
           - all
       use_launchable:
@@ -188,7 +193,7 @@ jobs:
           TEST_SUITE: ${{ inputs.test_suite }}
         run: |
           case "$TEST_SUITE" in
-            full|credential-sanitization|telegram-injection|messaging-providers|messaging-compatible-endpoint|dashboard-remote-bind|gpu|all) ;;
+            full|credential-sanitization|telegram-injection|messaging-providers|messaging-compatible-endpoint|dashboard-remote-bind|tool-disclosure-performance-smoke|gpu|all) ;;
             *)
               echo "::error::test_suite is not one of the supported Brev E2E suites"
               exit 1
@@ -213,7 +218,7 @@ jobs:
       - name: Checkout target branch
         uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
         with:
-          ref: ${{ env.RESOLVED_BRANCH || inputs.branch || 'main' }}
+          ref: ${{ env.RESOLVED_BRANCH || inputs.branch || github.ref }}
           persist-credentials: false
 
       - id: tested-ref
@@ -267,7 +272,28 @@ jobs:
       - name: Build CLI
         run: npm run build:cli
 
-      - name: Run ephemeral Brev E2E
+      - name: Install and verify Brev performance smoke tunnel prerequisite
+        if: inputs.test_suite == 'tool-disclosure-performance-smoke'
+        env:
+          CLOUDFLARED_VERSION: "2026.6.1"
+          CLOUDFLARED_DEB_SHA256: "ccd02ec216c62bfa573395d8f72cb2e91e95cbdf8726a8acc06b3e2d9aa31526"
+        run: |
+          set -euo pipefail
+          cloudflared_deb="${RUNNER_TEMP}/cloudflared-${CLOUDFLARED_VERSION}-linux-amd64.deb"
+          curl -fL "https://github.com/cloudflare/cloudflared/releases/download/${CLOUDFLARED_VERSION}/cloudflared-linux-amd64.deb" -o "${cloudflared_deb}"
+          printf '%s  %s\n' "${CLOUDFLARED_DEB_SHA256}" "${cloudflared_deb}" | sha256sum -c -
+          package="$(dpkg-deb -f "${cloudflared_deb}" Package)"
+          version="$(dpkg-deb -f "${cloudflared_deb}" Version)"
+          architecture="$(dpkg-deb -f "${cloudflared_deb}" Architecture)"
+          if [[ "${package}" != "cloudflared" || "${version}" != "${CLOUDFLARED_VERSION}" || "${architecture}" != "amd64" ]]; then
+            printf 'Unexpected cloudflared package metadata: package=%s version=%s architecture=%s\n' "${package}" "${version}" "${architecture}" >&2
+            exit 1
+          fi
+          sudo dpkg -i "${cloudflared_deb}"
+          cloudflared --version | grep -F "cloudflared version ${CLOUDFLARED_VERSION}"
+
+      - id: brev-test
+        name: Run ephemeral Brev E2E
         env:
           NEMOCLAW_RUN_BRANCH_VALIDATION_E2E: "1"
           NEMOCLAW_BREV_WORKFLOW_OWNS_INSTANCE: "1"
@@ -277,6 +303,9 @@ jobs:
           INSTANCE_NAME: ${{ env.BREV_E2E_INSTANCE_NAME }}
           TEST_SUITE: ${{ inputs.test_suite }}
           USE_LAUNCHABLE: ${{ inputs.use_launchable && '1' || '0' }}
+          BREV_MIN_VCPU: ${{ inputs.test_suite == 'tool-disclosure-performance-smoke' && '8' || '' }}
+          BREV_MIN_RAM: ${{ inputs.test_suite == 'tool-disclosure-performance-smoke' && '32' || '' }}
+          BREV_MIN_DISK: ${{ inputs.test_suite == 'tool-disclosure-performance-smoke' && '100' || '' }}
           BREV_PROVIDER: ${{ inputs.brev_provider || vars.BREV_PROVIDER || '' }}
           BREV_GPU_TYPE: ${{ inputs.brev_gpu_type || vars.BREV_GPU_TYPE || '' }}
           BREV_GPU_NAME: ${{ inputs.brev_gpu_name || vars.BREV_GPU_NAME || '' }}
@@ -286,6 +315,63 @@ jobs:
           KEEP_ALIVE: ${{ inputs.keep_alive }}
         run: npx vitest run --project e2e-branch-validation --silent=false --reporter=default
 
+      - name: Collect Brev tool-disclosure performance smoke artifacts
+        if: always() && inputs.test_suite == 'tool-disclosure-performance-smoke'
+        env:
+          INSTANCE: ${{ env.BREV_E2E_INSTANCE_NAME }}
+          TEST_OUTCOME: ${{ steps.brev-test.outcome }}
+        run: |
+          set -euo pipefail
+          artifact_dir="brev-tool-disclosure-performance-smoke-artifacts"
+          remote_dir="/tmp/nemoclaw-tool-disclosure-performance-smoke-artifacts"
+          archive="${RUNNER_TEMP}/brev-tool-disclosure-performance-smoke-artifacts.tar.gz"
+          mkdir -p "$artifact_dir"
+          brev refresh >/dev/null
+          if ! ssh -o StrictHostKeyChecking=no -o LogLevel=ERROR \
+               -o ConnectTimeout=10 "$INSTANCE" \
+               "test -d '$remote_dir' && tar -C '$remote_dir' -czf - ." \
+               >"$archive"; then
+            if [ "$TEST_OUTCOME" = "success" ]; then
+              echo "::error::Successful Brev performance smoke did not produce an artifact archive"
+              exit 1
+            fi
+            echo "::warning::Failed Brev performance smoke produced no artifact archive"
+            exit 0
+          fi
+          tar -C "$artifact_dir" -xzf "$archive"
+          result="$artifact_dir/tool-disclosure-hosted-inference-performance-smoke-completes-one-frozen-task-in-both-modes/tool-disclosure-performance-smoke.json"
+          if [ ! -f "$result" ]; then
+            if [ "$TEST_OUTCOME" = "success" ]; then
+              echo "::error::Successful Brev performance smoke artifact is missing its result JSON"
+              exit 1
+            fi
+            echo "::warning::Failed Brev performance smoke has only partial artifacts"
+          fi
+
+      - name: Scan Brev tool-disclosure performance smoke artifacts
+        if: always() && inputs.test_suite == 'tool-disclosure-performance-smoke'
+        env:
+          NVIDIA_INFERENCE_API_KEY: ${{ secrets.NVIDIA_INFERENCE_API_KEY }}
+        run: |
+          set -euo pipefail
+          artifact_dir="brev-tool-disclosure-performance-smoke-artifacts"
+          test -d "$artifact_dir"
+          if [ -n "${NVIDIA_INFERENCE_API_KEY:-}" ] && \
+             grep -R -I -l -F -- "$NVIDIA_INFERENCE_API_KEY" "$artifact_dir" >/dev/null; then
+            echo "::error::Brev performance smoke artifacts contain an unredacted inference credential"
+            exit 1
+          fi
+
+      - name: Upload Brev tool-disclosure performance smoke artifacts
+        if: always() && inputs.test_suite == 'tool-disclosure-performance-smoke'
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: e2e-brev-tool-disclosure-performance-smoke-${{ github.run_attempt }}
+          path: brev-tool-disclosure-performance-smoke-artifacts/
+          if-no-files-found: ignore
+          include-hidden-files: false
+          retention-days: 14
+
       # Collect debugging artifacts from the Brev VM on failure before the
       # instance gets torn down. Captures the onboard log, sandbox list,
       # docker state, and gateway status so downstream onboard failures are
@@ -306,6 +392,11 @@ jobs:
                timeout 15s openshell sandbox list  > /tmp/nc-debug/sandbox-list.txt 2>&1
                timeout 15s openshell gateway status > /tmp/nc-debug/gateway-status.txt 2>&1
                timeout 15s docker ps -a             > /tmp/nc-debug/docker-ps.txt 2>&1
+               free -h                              > /tmp/nc-debug/memory.txt 2>&1
+               df -h                                > /tmp/nc-debug/disk.txt 2>&1
+               timeout 15s sudo systemctl status docker --no-pager > /tmp/nc-debug/docker-service.txt 2>&1
+               timeout 15s sudo journalctl -u docker --since "15 minutes ago" --no-pager > /tmp/nc-debug/docker-journal.txt 2>&1
+               timeout 15s sudo dmesg | tail -200   > /tmp/nc-debug/kernel-tail.txt 2>&1
                timeout 30s tar -C /tmp -czf /tmp/nc-debug.tar.gz nc-debug
               ' || true
           scp -o StrictHostKeyChecking=no -o LogLevel=ERROR \

.github/workflows/e2e.yaml

@@ -14,7 +14,7 @@ on:
         default: ""
         type: string
       jobs:
-        description: "Optional comma-separated free-standing live E2E job ids. Empty runs default-enabled jobs only when targets is also empty; explicit-only jobs openshell-gateway-auth-contract, mcp-bridge-dev, hermes-gpu-startup, sandbox-rlimits-connect, and jetson-nvmap-gpu are skipped unless selected."
+        description: "Optional comma-separated free-standing live E2E job ids. Empty runs default-enabled jobs only when targets is also empty; explicit-only jobs openshell-gateway-auth-contract, mcp-bridge-dev, tool-disclosure-performance-smoke, hermes-gpu-startup, sandbox-rlimits-connect, and jetson-nvmap-gpu are skipped unless selected."
         required: false
         default: ""
         type: string
@@ -723,6 +723,96 @@ jobs:
         shell: bash
         run: bash .github/scripts/docker-auth-cleanup.sh
 
+  tool-disclosure-performance-smoke:
+    needs: generate-matrix
+    # This smoke provisions real agent sandboxes and consumes hosted inference.
+    # Keep it opt-in: it validates performance-test plumbing on ordinary CI hardware,
+    # but it is not the complete two-campaign performance-test procedure.
+    if: ${{ contains(format(',{0},', inputs.jobs), ',tool-disclosure-performance-smoke,') || contains(format(',{0},', inputs.targets), ',tool-disclosure-performance-smoke,') }}
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    timeout-minutes: 60
+    env:
+      E2E_JOB: "1"
+      E2E_DEFAULT_ENABLED: "0"
+      E2E_TARGET_ID: "tool-disclosure-performance-smoke"
+      E2E_ARTIFACT_DIR: ${{ github.workspace }}/e2e-artifacts/live/tool-disclosure-performance-smoke
+      NEMOCLAW_CLI_BIN: ${{ github.workspace }}/bin/nemoclaw.js
+      NEMOCLAW_RUN_LIVE_E2E: "1"
+      NEMOCLAW_E2E_USE_HOSTED_INFERENCE: "1"
+      NEMOCLAW_NON_INTERACTIVE: "1"
+      NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE: "1"
+      NEMOCLAW_OPENSHELL_CHANNEL: stable
+      NEMOCLAW_PROVIDER: custom
+      NEMOCLAW_ENDPOINT_URL: https://inference-api.nvidia.com/v1
+      NEMOCLAW_MODEL: nvidia/nvidia/nemotron-3-ultra
+      NEMOCLAW_COMPAT_MODEL: nvidia/nvidia/nemotron-3-ultra
+      NEMOCLAW_PREFERRED_API: openai-completions
+      OPENSHELL_GATEWAY: nemoclaw
+    steps:
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        with:
+          persist-credentials: false
+
+      - *dockerhub-auth
+
+      - name: Prepare E2E workspace
+        uses: NVIDIA/NemoClaw/.github/actions/prepare-e2e@50281ee84c4a6fc759da95ea28fc0b7d9c378a28
+
+      - name: Install and verify cloudflared prerequisite
+        env:
+          CLOUDFLARED_VERSION: "2026.6.1"
+          CLOUDFLARED_DEB_SHA256: "ccd02ec216c62bfa573395d8f72cb2e91e95cbdf8726a8acc06b3e2d9aa31526"
+        run: |
+          set -euo pipefail
+          cloudflared_deb="${RUNNER_TEMP}/cloudflared-${CLOUDFLARED_VERSION}-linux-amd64.deb"
+          curl -fL "https://github.com/cloudflare/cloudflared/releases/download/${CLOUDFLARED_VERSION}/cloudflared-linux-amd64.deb" -o "${cloudflared_deb}"
+          printf '%s  %s\n' "${CLOUDFLARED_DEB_SHA256}" "${cloudflared_deb}" | sha256sum -c -
+          package="$(dpkg-deb -f "${cloudflared_deb}" Package)"
+          version="$(dpkg-deb -f "${cloudflared_deb}" Version)"
+          architecture="$(dpkg-deb -f "${cloudflared_deb}" Architecture)"
+          if [[ "${package}" != "cloudflared" || "${version}" != "${CLOUDFLARED_VERSION}" || "${architecture}" != "amd64" ]]; then
+            printf 'Unexpected cloudflared package metadata: package=%s version=%s architecture=%s\n' "${package}" "${version}" "${architecture}" >&2
+            exit 1
+          fi
+          sudo dpkg -i "${cloudflared_deb}"
+          cloudflared --version | grep -F "cloudflared version ${CLOUDFLARED_VERSION}"
+
+      - name: Install OpenShell CLI
+        env:
+          NEMOCLAW_OPENSHELL_FORCE_INSTALL: "1"
+        run: bash scripts/install-openshell.sh
+
+      - name: Run tool-disclosure performance smoke live test
+        env:
+          NVIDIA_INFERENCE_API_KEY: ${{ secrets.NVIDIA_INFERENCE_API_KEY }}
+        run: |
+          set -euo pipefail
+          export PATH="$HOME/.local/bin:$HOME/.npm-global/bin:$PATH"
+          if command -v openshell >/dev/null 2>&1; then
+            OPENSHELL_BIN="$(command -v openshell)"
+          elif [ -x "$HOME/.local/bin/openshell" ]; then
+            OPENSHELL_BIN="$HOME/.local/bin/openshell"
+          else
+            echo "::error::OpenShell CLI not found after install"
+            exit 1
+          fi
+          export OPENSHELL_BIN
+          "$OPENSHELL_BIN" --version
+          npx vitest run --project e2e-live \
+            test/e2e/live/tool-disclosure-performance-smoke.test.ts \
+            --silent=false --reporter=default
+
+      - name: Upload tool-disclosure performance smoke artifacts
+        if: always()
+        uses: NVIDIA/NemoClaw/.github/actions/upload-e2e-artifacts@7768e15eb90d3ee2d33432f481dfe8747e4f6d57
+
+      - name: Clean up Docker auth
+        if: always()
+        shell: bash
+        run: bash .github/scripts/docker-auth-cleanup.sh
+
   onboard-negative-paths:
     needs: generate-matrix
     if: ${{ (github.event_name != 'workflow_dispatch' || (inputs.jobs == '' && inputs.targets == '')) || contains(format(',{0},', inputs.jobs), ',onboard-negative-paths,') || contains(format(',{0},', inputs.targets), ',onboard-negative-paths,') }}
@@ -4613,6 +4703,7 @@ jobs:
         openshell-gateway-auth-contract,
         mcp-bridge,
         mcp-bridge-dev,
+        tool-disclosure-performance-smoke,
         onboard-negative-paths,
         skill-agent,
         openclaw-skill-cli,
@@ -4718,6 +4809,11 @@ jobs:
                 target: 'mcp-bridge-dev',
                 reason: 'default dispatch excludes moving OpenShell dev artifacts unless explicitly selected',
               },
+              'tool-disclosure-performance-smoke': {
+                job: 'tool-disclosure-performance-smoke',
+                target: 'tool-disclosure-performance-smoke',
+                reason: 'default dispatch excludes the hosted-inference performance smoke test unless explicitly selected',
+              },
               'jetson-nvmap-gpu': {
                 job: 'jetson-nvmap-gpu',
                 target: 'jetson-nvmap-gpu',
@@ -4848,7 +4944,7 @@ jobs:
                 ? '**Requested jobs:** _(selector rejected by workflow validation)_'
                 : requestedJobs
                   ? `**Requested jobs:** \`${requestedJobs}\``
-                  : '**Requested jobs:** _(default — all default-enabled free-standing jobs; explicit-only jobs `openshell-gateway-auth-contract`, `mcp-bridge-dev`, `hermes-gpu-startup`, `sandbox-rlimits-connect`, and `jetson-nvmap-gpu` are skipped unless selected)_',
+                  : '**Requested jobs:** _(default — all default-enabled free-standing jobs; explicit-only jobs `openshell-gateway-auth-contract`, `mcp-bridge-dev`, `tool-disclosure-performance-smoke`, `hermes-gpu-startup`, `sandbox-rlimits-connect`, and `jetson-nvmap-gpu` are skipped unless selected)_',
               `**Summary:** ${passed.length} passed, ${failed.length} failed, ${cancelled.length} cancelled, ${skipped.length} skipped`,
               '',
               '| Job | Result |',

agents/langchain-deepagents-code/Dockerfile

@@ -59,14 +59,16 @@ RUN case "$NEMOCLAW_TOOL_DISCLOSURE" in \
     esac
 
 # The launcher and startup script read these root-owned files instead of
-# trusting process-level environment overrides for inference routing. Invoking
-# each launcher validates the build args before the image can complete.
+# trusting process-level environment overrides for inference routing or tool
+# disclosure. Invoking each launcher validates the build args before the image
+# can complete.
 RUN install -d -m 0755 /usr/local/share/nemoclaw \
     && printf '%s\n' "$NEMOCLAW_PROXY_HOST" > /usr/local/share/nemoclaw/dcode-proxy-host \
     && printf '%s\n' "$NEMOCLAW_PROXY_PORT" > /usr/local/share/nemoclaw/dcode-proxy-port \
+    && printf '%s\n' "$NEMOCLAW_TOOL_DISCLOSURE" > /usr/local/share/nemoclaw/dcode-tool-disclosure \
     && printf '%s\n' "$NEMOCLAW_INFERENCE_BASE_URL" > /usr/local/share/nemoclaw/dcode-inference-base-url \
-    && chown root:root /usr/local/share/nemoclaw/dcode-proxy-host /usr/local/share/nemoclaw/dcode-proxy-port /usr/local/share/nemoclaw/dcode-inference-base-url \
-    && chmod 0444 /usr/local/share/nemoclaw/dcode-proxy-host /usr/local/share/nemoclaw/dcode-proxy-port /usr/local/share/nemoclaw/dcode-inference-base-url \
+    && chown root:root /usr/local/share/nemoclaw/dcode-proxy-host /usr/local/share/nemoclaw/dcode-proxy-port /usr/local/share/nemoclaw/dcode-tool-disclosure /usr/local/share/nemoclaw/dcode-inference-base-url \
+    && chmod 0444 /usr/local/share/nemoclaw/dcode-proxy-host /usr/local/share/nemoclaw/dcode-proxy-port /usr/local/share/nemoclaw/dcode-tool-disclosure /usr/local/share/nemoclaw/dcode-inference-base-url \
     && /usr/local/bin/dcode --version \
     && /usr/local/bin/dcode.real --version \
     && /usr/local/bin/deepagents-code --version

agents/langchain-deepagents-code/Dockerfile.base

@@ -42,7 +42,7 @@ RUN groupadd -r sandbox \
 
 # Pre-create shell init files for the sandbox user. OpenShell/NemoClaw writes
 # /tmp/nemoclaw-proxy-env.sh at startup so interactive sessions and dcode share
-# the same proxy, CA, inference, HOME, and update-check posture.
+# the same proxy, CA, inference, tool-disclosure, HOME, and update-check posture.
 # hadolint ignore=SC2016
 RUN printf '%s\n' \
         '# Source runtime proxy + Deep Agents Code config' \