ci: enforce strict zizmor audits and harden workflows / 启用严格 zizmor 审计并加固工作流

.github/dependabot.yml

@@ -8,3 +8,5 @@ updates:
     github-actions:
       patterns:
         - "*"
+  cooldown:
+    default-days: 7

.github/workflows/benchmark-multinode-tmpl.yml

@@ -2,6 +2,15 @@ name: Template - Multi-Node Benchmark
 
 on:
   workflow_call:
+    secrets:
+      REPO_PAT:
+        required: true
+      INFERENCEX_OFFICIAL_RO_HF_TOKEN:
+        required: true
+      MODAL_TOKEN_ID:
+        required: false
+      MODAL_TOKEN_SECRET:
+        required: false
     inputs:
       config:
         description: "Benchmark configuration as JSON"
@@ -251,7 +260,7 @@ jobs:
       - name: Slurm cleanup (pre-run)
         run: &slurm-cleanup |
           if command -v squeue >/dev/null 2>&1; then
-            for job_name in "${{ runner.name }}" "inferencex-${{ runner.name }}"; do
+            for job_name in "${RUNNER_NAME}" "inferencex-${RUNNER_NAME}"; do
               echo "[Slurm] Cleaning up jobs named: $job_name ..."
               scancel --user="$USER" --name="$job_name" || true
               while [ -n "$(squeue --user="$USER" --name="$job_name" --noheader --format='%i')" ]; do
@@ -272,7 +281,7 @@ jobs:
           # module git dir — a truncated clone may lack HEAD entirely, and a
           # git dir whose HEAD does not resolve to a commit is dropped
           # together with its worktree so checkout re-clones it.
-          repo_dir="${{ github.workspace }}"
+          repo_dir="${GITHUB_WORKSPACE}"
           if [ -d "${repo_dir}/.git" ]; then
             find "${repo_dir}/.git" -name index.lock -type f -delete || true
             for gitdir in "${repo_dir}"/.git/modules/* "${repo_dir}"/.git/modules/*/*; do
@@ -293,9 +302,12 @@ jobs:
           ref: ${{ inputs.ref || github.sha }}
           clean: true
           submodules: true
+          persist-credentials: false
 
       - name: Launch multi-node job script
         env:
+          PREFILL_ADDITIONAL_SETTINGS: ${{ toJSON(fromJSON(inputs.config).prefill.additional-settings) }}
+          DECODE_ADDITIONAL_SETTINGS: ${{ toJSON(fromJSON(inputs.config).decode.additional-settings) }}
           RUNNER_NAME: ${{ runner.name }}
           RUNNER_TYPE: ${{ inputs.runner }}
           RESULT_FILENAME_BASE: ${{ env.EXP_NAME }}_${{ env.PRECISION }}_${{ env.FRAMEWORK }}_prefill-tp${{ env.PREFILL_TP }}-pp${{ env.PREFILL_PP_SIZE }}-dcp${{ env.PREFILL_DCP_SIZE }}-pcp${{ env.PREFILL_PCP_SIZE }}-ep${{ env.PREFILL_EP }}-dp${{ env.PREFILL_DP_ATTN }}-nw${{ env.PREFILL_NUM_WORKERS }}_decode-tp${{ env.DECODE_TP }}-pp${{ env.DECODE_PP_SIZE }}-dcp${{ env.DECODE_DCP_SIZE }}-pcp${{ env.DECODE_PCP_SIZE }}-ep${{ env.DECODE_EP }}-dp${{ env.DECODE_DP_ATTN }}-nw${{ env.DECODE_NUM_WORKERS }}_disagg-${{ env.DISAGG }}_spec-${{ env.SPEC_DECODING }}_conc${{ join(fromJson(inputs.conc-list), 'x') }}_${{ runner.name }}
@@ -326,22 +338,31 @@ jobs:
           if [[ -f benchmarks/multi_node/runtime_settings.sh ]]; then
             source benchmarks/multi_node/runtime_settings.sh
           fi
-          export ${{ join(fromJSON(inputs.config).prefill.additional-settings, ' ') }} ${{ join(fromJSON(inputs.config).decode.additional-settings, ' ') }}
+          # Assign data literally; recipe values must never become shell code.
+          settings_json=$(jq -cen \
+            --argjson prefill "$PREFILL_ADDITIONAL_SETTINGS" \
+            --argjson decode "$DECODE_ADDITIONAL_SETTINGS" \
+            '($prefill // []) + ($decode // []) |
+             if all(.[]; type == "string" and test("^[A-Za-z_][A-Za-z0-9_]*=")) then .
+             else error("additional-settings must contain NAME=value assignments") end')
+          while IFS= read -r -d '' setting; do
+            export "$setting"
+          done < <(jq -j '.[] + "\u0000"' <<< "$settings_json")
           # Resolve the workflow's documented automatic eval-concurrency selection.
           if [[ -z "$EVAL_CONC" ]]; then
             EVAL_CONC=$(python3 -c 'import os; print(max(map(int, os.environ["CONC_LIST"].split())))')
           fi
           export EVAL_CONC
           export IS_MULTINODE=true
-          bash ./runners/launch_${RUNNER_NAME%%_*}.sh
-          if [ "${{ inputs.eval-only }}" = "true" ]; then
+          bash "./runners/launch_${RUNNER_NAME%%_*}.sh"
+          if [ "${EVAL_ONLY}" = "true" ]; then
             echo "Eval-only mode: skipping benchmark result file check"
             # Verify eval produced results
             if ! ls results*.json 1>/dev/null 2>&1; then
               echo "Eval-only run failed: no results*.json files found." >&2
               exit 1
             fi
-          elif [ "${{ inputs.scenario-type }}" = "agentic-coding" ]; then
+          elif [ "${SCENARIO_TYPE}" = "agentic-coding" ]; then
             expected_count=$(wc -w <<< "$CONC_LIST" | tr -d ' ')
             shopt -s nullglob
             agentic_results=("${RESULT_FILENAME}"_conc*.json)

.github/workflows/benchmark-tmpl.yml

@@ -1,6 +1,15 @@
 name: Template - Benchmark
 on:
   workflow_call:
+    secrets:
+      REPO_PAT:
+        required: true
+      INFERENCEX_OFFICIAL_RO_HF_TOKEN:
+        required: true
+      MODAL_TOKEN_ID:
+        required: false
+      MODAL_TOKEN_SECRET:
+        required: false
     inputs:
       config:
         description: "Benchmark configuration as JSON"
@@ -163,15 +172,15 @@ jobs:
             format(
               '["self-hosted",{0},{1},{2},{3},{4}]',
               toJSON(inputs.runner),
-              toJSON('nodes:1'),
+              '"nodes:1"',
               toJSON(format('ci-job-{0}-{1}', inputs.priority, inputs.queue-token)),
               toJSON(format('ci-attempt-{0}', github.run_attempt)),
               toJSON(format('ci-skip-queue-pr-{0}', inputs.skip-queue-pr))
             ) ||
             format(
               '["self-hosted",{0},{1},{2},{3}]',
               toJSON(inputs.runner),
-              toJSON('nodes:1'),
+              '"nodes:1"',
               toJSON(format('ci-job-{0}-{1}', inputs.priority, inputs.queue-token)),
               toJSON(format('ci-attempt-{0}', github.run_attempt))
             )
@@ -204,15 +213,15 @@ jobs:
       ${{ inputs.scenario-type == 'agentic-coding' && fromJSON(inputs.config).kv-offload-backend.name != 'none' && fromJSON(inputs.config).kv-offload-backend.name != 'default' && format('{0}', fromJSON(inputs.config).kv-offload-backend.name) || '' }}
       c${{ fromJSON(inputs.config).conc }}${{ inputs.eval-only && ' | eval-only' || (inputs.run-eval && ' | eval' || '') }}
     steps:
-      - name: Resource cleanup (pre-run)
-        run: &resource-cleanup |
-          # Containerized AMD Slurm jobs can leave root-owned artifacts when
-          # interrupted before their launcher EXIT trap runs. Repair ownership
-          # before checkout cleanup and again after the job via this shared step.
-          if [[ "${{ inputs.runner }}" == "cluster:mi355x-amds" && -d "$GITHUB_WORKSPACE" ]]; then
+      - name: Repair root-owned artifacts (pre-run)
+        if: ${{ inputs.runner == 'cluster:mi355x-amds' }}
+        run: |
+          if [[ -d "$GITHUB_WORKSPACE" ]]; then
             sudo chown -R "$(id -u):$(id -g)" "$GITHUB_WORKSPACE"
           fi
 
+      - name: Resource cleanup (pre-run)
+        run: &resource-cleanup |
           # Cleanup Docker resources
           if command -v docker >/dev/null 2>&1 && docker info >/dev/null 2>&1; then
             echo "[Docker] Cleaning up resources ..."
@@ -226,10 +235,10 @@ jobs:
 
           # Cleanup SLURM resources
           if command -v squeue >/dev/null 2>&1; then
-            echo "[Slurm] Cleaning up jobs with name: ${{ runner.name }} ..."
-            scancel --name="${{ runner.name }}" || true
-            while [ -n "$(squeue --name='${{ runner.name }}' --noheader --format='%i')" ]; do
-              squeue --name="${{ runner.name }}"
+            echo "[Slurm] Cleaning up jobs with name: ${RUNNER_NAME} ..."
+            scancel --name="${RUNNER_NAME}" || true
+            while [ -n "$(squeue --name="${RUNNER_NAME}" --noheader --format='%i')" ]; do
+              squeue --name="${RUNNER_NAME}"
               sleep 5
             done
           fi
@@ -245,7 +254,7 @@ jobs:
           # module git dir — a truncated clone may lack HEAD entirely, and a
           # git dir whose HEAD does not resolve to a commit is dropped
           # together with its worktree so checkout re-clones it.
-          repo_dir="${{ github.workspace }}"
+          repo_dir="${GITHUB_WORKSPACE}"
           if [ -d "${repo_dir}/.git" ]; then
             find "${repo_dir}/.git" -name index.lock -type f -delete || true
             for gitdir in "${repo_dir}"/.git/modules/* "${repo_dir}"/.git/modules/*/*; do
@@ -266,6 +275,7 @@ jobs:
           ref: ${{ inputs.ref || github.sha }}
           clean: true
           submodules: true
+          persist-credentials: false
 
       - name: Launch job script
         env:
@@ -297,9 +307,9 @@ jobs:
           if [[ -f runners/runtime_settings.sh ]]; then
             source runners/runtime_settings.sh
           fi
-          bash ./runners/launch_${RUNNER_NAME%%_*}.sh
+          bash "./runners/launch_${RUNNER_NAME%%_*}.sh"
 
-          if [ "${{ inputs.eval-only }}" = "true" ]; then
+          if [ "${EVAL_ONLY}" = "true" ]; then
             echo "Eval-only mode: skipping benchmark result file check"
             # Verify eval produced results
             if ! ls results*.json 1>/dev/null 2>&1; then
@@ -322,7 +332,7 @@ jobs:
               exit 1
             fi
 
-            if [ "${{ inputs.scenario-type }}" = "agentic-coding" ]; then
+            if [ "${SCENARIO_TYPE}" = "agentic-coding" ]; then
               python3 -m utils.agentic.validation.validate_agentic_result \
                 results/aiperf_artifacts \
                 --failed-request-threshold "$AIPERF_FAILED_REQUEST_THRESHOLD"
@@ -451,6 +461,13 @@ jobs:
           rm -f -- ./*_artifacts.tar.gz || true
           rm -f agent_preds.json predictions.jsonl swebench_report_*.json *.traj* || true
 
+      - name: Repair root-owned artifacts (post-run)
+        if: ${{ always() && inputs.runner == 'cluster:mi355x-amds' }}
+        run: |
+          if [[ -d "$GITHUB_WORKSPACE" ]]; then
+            sudo chown -R "$(id -u):$(id -g)" "$GITHUB_WORKSPACE"
+          fi
+
       - name: Resource cleanup (post-run)
         if: always()
         run: *resource-cleanup

.github/workflows/ci.yml

@@ -3,10 +3,15 @@ name: CI
 on:
   pull_request:
     types: [opened, synchronize, reopened, ready_for_review]
-    paths: ['**/*.py']
+    paths: &python-paths
+      - '**/*.py'
+      - '.github/workflows/ci.yml'
+      - '.claude/requirements-mcp.txt'
+      - 'infx/ruff.toml'
+      - '**/pytest.ini'
   push:
     branches: [main]
-    paths: ['**/*.py']
+    paths: *python-paths
   workflow_dispatch:
 
 permissions: