fix: merge multiple --mcp-config flags and support --allowed-tools parsing (#748 )

* fix: merge multiple --mcp-config flags instead of overwriting When users provide their own --mcp-config in claude_args, the action's built-in MCP servers (github_comment, github_ci, etc.) were being lost because multiple --mcp-config flags were overwriting each other. This fix: - Adds mcp-config to ACCUMULATING_FLAGS to collect all values - Changes delimiter to null character to avoid conflicts with JSON - Adds mergeMcpConfigs() to combine mcpServers objects from multiple configs - Merges inline JSON configs while preserving file path configs Fixes #745 🤖 Generated with [Claude Code](https://claude.ai/code) Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Ashwin Bhat <ashwin-ant@users.noreply.github.com> * fix: support hyphenated --allowed-tools flag and multiple values The --allowed-tools flag was not being parsed correctly when: 1. Using the hyphenated form (--allowed-tools) instead of camelCase (--allowedTools) 2. Passing multiple space-separated values after a single flag (e.g., --allowed-tools "Tool1" "Tool2" "Tool3") This fix: - Adds hyphenated variants (allowed-tools, disallowed-tools) to ACCUMULATING_FLAGS - Updates parsing to consume all consecutive non-flag values for accumulating flags - Merges values from both camelCase and hyphenated variants Fixes #746 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Ashwin Bhat <ashwin-ant@users.noreply.github.com>
chore: update model to claude-opus-4-5 in workflow (#747 )
2026-01-23 15:04:13 +08:00 · 2025-12-16 13:08:25 -08:00 · 2025-12-16 12:47:41 -08:00 · 2025-12-15 23:53:03 +00:00 · 2025-12-14 19:42:54 -08:00 · 2025-12-13 01:00:43 +00:00
28 changed files with 1818 additions and 43 deletions
--- a/.github/workflows/claude.yml
+++ b/.github/workflows/claude.yml
@@ -36,4 +36,4 @@ jobs:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          claude_args: |
            --allowedTools "Bash(bun install),Bash(bun test:*),Bash(bun run format),Bash(bun typecheck)"
-            --model "claude-opus-4-1-20250805"
+            --model "claude-opus-4-5"
--- a/.github/workflows/sync-base-action.yml
+++ b/.github/workflows/sync-base-action.yml
@@ -94,5 +94,5 @@ jobs:
          echo "✅ Successfully synced \`base-action\` directory to [anthropics/claude-code-base-action](https://github.com/anthropics/claude-code-base-action)" >> $GITHUB_STEP_SUMMARY
          echo "" >> $GITHUB_STEP_SUMMARY
          echo "- **Source commit**: [\`${GITHUB_SHA:0:7}\`](https://github.com/anthropics/claude-code-action/commit/${GITHUB_SHA})" >> $GITHUB_STEP_SUMMARY
-          echo "- **Triggered by**: ${{ github.event_name }}" >> $GITHUB_STEP_SUMMARY
-          echo "- **Actor**: @${{ github.actor }}" >> $GITHUB_STEP_SUMMARY
+          echo "- **Triggered by**: $GITHUB_EVENT_NAME" >> $GITHUB_STEP_SUMMARY
+          echo "- **Actor**: @$GITHUB_ACTOR" >> $GITHUB_STEP_SUMMARY
--- a/.github/workflows/test-base-action.yml
+++ b/.github/workflows/test-base-action.yml
@@ -118,3 +118,61 @@ jobs:
            echo "❌ Execution log file not found"
            exit 1
          fi
+
+  test-agent-sdk:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4
+
+      - name: Test with Agent SDK
+        id: sdk-test
+        uses: ./base-action
+        env:
+          USE_AGENT_SDK: "true"
+        with:
+          prompt: ${{ github.event.inputs.test_prompt || 'List the files in the current directory starting with "package"' }}
+          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
+          allowed_tools: "LS,Read"
+
+      - name: Verify SDK output
+        run: |
+          OUTPUT_FILE="${{ steps.sdk-test.outputs.execution_file }}"
+          CONCLUSION="${{ steps.sdk-test.outputs.conclusion }}"
+
+          echo "Conclusion: $CONCLUSION"
+          echo "Output file: $OUTPUT_FILE"
+
+          if [ "$CONCLUSION" = "success" ]; then
+            echo "✅ Action completed successfully with Agent SDK"
+          else
+            echo "❌ Action failed with Agent SDK"
+            exit 1
+          fi
+
+          if [ -f "$OUTPUT_FILE" ]; then
+            if [ -s "$OUTPUT_FILE" ]; then
+              echo "✅ Execution log file created successfully with content"
+              echo "Validating JSON format:"
+              if jq . "$OUTPUT_FILE" > /dev/null 2>&1; then
+                echo "✅ Output is valid JSON"
+                # Verify SDK output contains total_cost_usd (SDK field name)
+                if jq -e '.[] | select(.type == "result") | .total_cost_usd' "$OUTPUT_FILE" > /dev/null 2>&1; then
+                  echo "✅ SDK output contains total_cost_usd field"
+                else
+                  echo "❌ SDK output missing total_cost_usd field"
+                  exit 1
+                fi
+                echo "Content preview:"
+                head -c 500 "$OUTPUT_FILE"
+              else
+                echo "❌ Output is not valid JSON"
+                exit 1
+              fi
+            else
+              echo "❌ Execution log file is empty"
+              exit 1
+            fi
+          else
+            echo "❌ Execution log file not found"
+            exit 1
+          fi
--- a/action.yml
+++ b/action.yml
@@ -127,6 +127,9 @@ outputs:
  structured_output:
    description: "JSON string containing all structured output fields when --json-schema is provided in claude_args. Use fromJSON() to parse: fromJSON(steps.id.outputs.structured_output).field_name"
    value: ${{ steps.claude-code.outputs.structured_output }}
+  session_id:
+    description: "The Claude Code session ID that can be used with --resume to continue this conversation"
+    value: ${{ steps.claude-code.outputs.session_id }}

 runs:
  using: "composite"
@@ -140,10 +143,12 @@ runs:
    - name: Setup Custom Bun Path
      if: inputs.path_to_bun_executable != ''
      shell: bash
+      env:
+        PATH_TO_BUN_EXECUTABLE: ${{ inputs.path_to_bun_executable }}
      run: |
-        echo "Using custom Bun executable: ${{ inputs.path_to_bun_executable }}"
+        echo "Using custom Bun executable: $PATH_TO_BUN_EXECUTABLE"
        # Add the directory containing the custom executable to PATH
-        BUN_DIR=$(dirname "${{ inputs.path_to_bun_executable }}")
+        BUN_DIR=$(dirname "$PATH_TO_BUN_EXECUTABLE")
        echo "$BUN_DIR" >> "$GITHUB_PATH"

    - name: Install Dependencies
@@ -182,6 +187,8 @@ runs:
    - name: Install Base Action Dependencies
      if: steps.prepare.outputs.contains_trigger == 'true'
      shell: bash
+      env:
+        PATH_TO_CLAUDE_CODE_EXECUTABLE: ${{ inputs.path_to_claude_code_executable }}
      run: |
        echo "Installing base-action dependencies..."
        cd ${GITHUB_ACTION_PATH}/base-action
@@ -190,8 +197,8 @@ runs:
        cd -

        # Install Claude Code if no custom executable is provided
-        if [ -z "${{ inputs.path_to_claude_code_executable }}" ]; then
-          CLAUDE_CODE_VERSION="2.0.50"
+        if [ -z "$PATH_TO_CLAUDE_CODE_EXECUTABLE" ]; then
+          CLAUDE_CODE_VERSION="2.0.70"
          echo "Installing Claude Code v${CLAUDE_CODE_VERSION}..."
          for attempt in 1 2 3; do
            echo "Installation attempt $attempt..."
@@ -210,9 +217,9 @@ runs:
          echo "Claude Code installed successfully"
          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
        else
-          echo "Using custom Claude Code executable: ${{ inputs.path_to_claude_code_executable }}"
+          echo "Using custom Claude Code executable: $PATH_TO_CLAUDE_CODE_EXECUTABLE"
          # Add the directory containing the custom executable to PATH
-          CLAUDE_DIR=$(dirname "${{ inputs.path_to_claude_code_executable }}")
+          CLAUDE_DIR=$(dirname "$PATH_TO_CLAUDE_CODE_EXECUTABLE")
          echo "$CLAUDE_DIR" >> "$GITHUB_PATH"
        fi

--- a/base-action/action.yml
+++ b/base-action/action.yml
@@ -82,6 +82,9 @@ outputs:
  structured_output:
    description: "JSON string containing all structured output fields when --json-schema is provided in claude_args (use fromJSON() or jq to parse)"
    value: ${{ steps.run_claude.outputs.structured_output }}
+  session_id:
+    description: "The Claude Code session ID that can be used with --resume to continue this conversation"
+    value: ${{ steps.run_claude.outputs.session_id }}

 runs:
  using: "composite"
@@ -101,10 +104,12 @@ runs:
    - name: Setup Custom Bun Path
      if: inputs.path_to_bun_executable != ''
      shell: bash
+      env:
+        PATH_TO_BUN_EXECUTABLE: ${{ inputs.path_to_bun_executable }}
      run: |
-        echo "Using custom Bun executable: ${{ inputs.path_to_bun_executable }}"
+        echo "Using custom Bun executable: $PATH_TO_BUN_EXECUTABLE"
        # Add the directory containing the custom executable to PATH
-        BUN_DIR=$(dirname "${{ inputs.path_to_bun_executable }}")
+        BUN_DIR=$(dirname "$PATH_TO_BUN_EXECUTABLE")
        echo "$BUN_DIR" >> "$GITHUB_PATH"

    - name: Install Dependencies
@@ -115,9 +120,11 @@ runs:

    - name: Install Claude Code
      shell: bash
+      env:
+        PATH_TO_CLAUDE_CODE_EXECUTABLE: ${{ inputs.path_to_claude_code_executable }}
      run: |
-        if [ -z "${{ inputs.path_to_claude_code_executable }}" ]; then
-          CLAUDE_CODE_VERSION="2.0.50"
+        if [ -z "$PATH_TO_CLAUDE_CODE_EXECUTABLE" ]; then
+          CLAUDE_CODE_VERSION="2.0.70"
          echo "Installing Claude Code v${CLAUDE_CODE_VERSION}..."
          for attempt in 1 2 3; do
            echo "Installation attempt $attempt..."
@@ -135,9 +142,9 @@ runs:
          done
          echo "Claude Code installed successfully"
        else
-          echo "Using custom Claude Code executable: ${{ inputs.path_to_claude_code_executable }}"
+          echo "Using custom Claude Code executable: $PATH_TO_CLAUDE_CODE_EXECUTABLE"
          # Add the directory containing the custom executable to PATH
-          CLAUDE_DIR=$(dirname "${{ inputs.path_to_claude_code_executable }}")
+          CLAUDE_DIR=$(dirname "$PATH_TO_CLAUDE_CODE_EXECUTABLE")
          echo "$CLAUDE_DIR" >> "$GITHUB_PATH"
        fi

--- a/base-action/bun.lock
+++ b/base-action/bun.lock
@@ -1,10 +1,12 @@
 {
  "lockfileVersion": 1,
+  "configVersion": 0,
  "workspaces": {
    "": {
      "name": "@anthropic-ai/claude-code-base-action",
      "dependencies": {
        "@actions/core": "^1.10.1",
+        "@anthropic-ai/claude-agent-sdk": "^0.1.70",
        "shell-quote": "^1.8.3",
      },
      "devDependencies": {
@@ -25,8 +27,40 @@

    "@actions/io": ["@actions/io@1.1.3", "", {}, "sha512-wi9JjgKLYS7U/z8PPbco+PvTb/nRWjeoFlJ1Qer83k/3C5PHQi28hiVdeE2kHXmIL99mQFawx8qt/JPjZilJ8Q=="],

+    "@anthropic-ai/claude-agent-sdk": ["@anthropic-ai/claude-agent-sdk@0.1.70", "", { "optionalDependencies": { "@img/sharp-darwin-arm64": "^0.33.5", "@img/sharp-darwin-x64": "^0.33.5", "@img/sharp-linux-arm": "^0.33.5", "@img/sharp-linux-arm64": "^0.33.5", "@img/sharp-linux-x64": "^0.33.5", "@img/sharp-linuxmusl-arm64": "^0.33.5", "@img/sharp-linuxmusl-x64": "^0.33.5", "@img/sharp-win32-x64": "^0.33.5" }, "peerDependencies": { "zod": "^3.24.1" } }, "sha512-4jpFPDX8asys6skO1r3Pzh0Fe9nbND2ASYTWuyFB5iN9bWEL6WScTFyGokjql3M2TkEp9ZGuB2YYpTCdaqT9Sw=="],
+
    "@fastify/busboy": ["@fastify/busboy@2.1.1", "", {}, "sha512-vBZP4NlzfOlerQTnba4aqZoMhE/a9HY7HRqoOPaETQcSQuWEIyZMHGfVu6w9wGtGK5fED5qRs2DteVCjOH60sA=="],

+    "@img/sharp-darwin-arm64": ["@img/sharp-darwin-arm64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-darwin-arm64": "1.0.4" }, "os": "darwin", "cpu": "arm64" }, "sha512-UT4p+iz/2H4twwAoLCqfA9UH5pI6DggwKEGuaPy7nCVQ8ZsiY5PIcrRvD1DzuY3qYL07NtIQcWnBSY/heikIFQ=="],
+
+    "@img/sharp-darwin-x64": ["@img/sharp-darwin-x64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-darwin-x64": "1.0.4" }, "os": "darwin", "cpu": "x64" }, "sha512-fyHac4jIc1ANYGRDxtiqelIbdWkIuQaI84Mv45KvGRRxSAa7o7d1ZKAOBaYbnepLC1WqxfpimdeWfvqqSGwR2Q=="],
+
+    "@img/sharp-libvips-darwin-arm64": ["@img/sharp-libvips-darwin-arm64@1.0.4", "", { "os": "darwin", "cpu": "arm64" }, "sha512-XblONe153h0O2zuFfTAbQYAX2JhYmDHeWikp1LM9Hul9gVPjFY427k6dFEcOL72O01QxQsWi761svJ/ev9xEDg=="],
+
+    "@img/sharp-libvips-darwin-x64": ["@img/sharp-libvips-darwin-x64@1.0.4", "", { "os": "darwin", "cpu": "x64" }, "sha512-xnGR8YuZYfJGmWPvmlunFaWJsb9T/AO2ykoP3Fz/0X5XV2aoYBPkX6xqCQvUTKKiLddarLaxpzNe+b1hjeWHAQ=="],
+
+    "@img/sharp-libvips-linux-arm": ["@img/sharp-libvips-linux-arm@1.0.5", "", { "os": "linux", "cpu": "arm" }, "sha512-gvcC4ACAOPRNATg/ov8/MnbxFDJqf/pDePbBnuBDcjsI8PssmjoKMAz4LtLaVi+OnSb5FK/yIOamqDwGmXW32g=="],
+
+    "@img/sharp-libvips-linux-arm64": ["@img/sharp-libvips-linux-arm64@1.0.4", "", { "os": "linux", "cpu": "arm64" }, "sha512-9B+taZ8DlyyqzZQnoeIvDVR/2F4EbMepXMc/NdVbkzsJbzkUjhXv/70GQJ7tdLA4YJgNP25zukcxpX2/SueNrA=="],
+
+    "@img/sharp-libvips-linux-x64": ["@img/sharp-libvips-linux-x64@1.0.4", "", { "os": "linux", "cpu": "x64" }, "sha512-MmWmQ3iPFZr0Iev+BAgVMb3ZyC4KeFc3jFxnNbEPas60e1cIfevbtuyf9nDGIzOaW9PdnDciJm+wFFaTlj5xYw=="],
+
+    "@img/sharp-libvips-linuxmusl-arm64": ["@img/sharp-libvips-linuxmusl-arm64@1.0.4", "", { "os": "linux", "cpu": "arm64" }, "sha512-9Ti+BbTYDcsbp4wfYib8Ctm1ilkugkA/uscUn6UXK1ldpC1JjiXbLfFZtRlBhjPZ5o1NCLiDbg8fhUPKStHoTA=="],
+
+    "@img/sharp-libvips-linuxmusl-x64": ["@img/sharp-libvips-linuxmusl-x64@1.0.4", "", { "os": "linux", "cpu": "x64" }, "sha512-viYN1KX9m+/hGkJtvYYp+CCLgnJXwiQB39damAO7WMdKWlIhmYTfHjwSbQeUK/20vY154mwezd9HflVFM1wVSw=="],
+
+    "@img/sharp-linux-arm": ["@img/sharp-linux-arm@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linux-arm": "1.0.5" }, "os": "linux", "cpu": "arm" }, "sha512-JTS1eldqZbJxjvKaAkxhZmBqPRGmxgu+qFKSInv8moZ2AmT5Yib3EQ1c6gp493HvrvV8QgdOXdyaIBrhvFhBMQ=="],
+
+    "@img/sharp-linux-arm64": ["@img/sharp-linux-arm64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linux-arm64": "1.0.4" }, "os": "linux", "cpu": "arm64" }, "sha512-JMVv+AMRyGOHtO1RFBiJy/MBsgz0x4AWrT6QoEVVTyh1E39TrCUpTRI7mx9VksGX4awWASxqCYLCV4wBZHAYxA=="],
+
+    "@img/sharp-linux-x64": ["@img/sharp-linux-x64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linux-x64": "1.0.4" }, "os": "linux", "cpu": "x64" }, "sha512-opC+Ok5pRNAzuvq1AG0ar+1owsu842/Ab+4qvU879ippJBHvyY5n2mxF1izXqkPYlGuP/M556uh53jRLJmzTWA=="],
+
+    "@img/sharp-linuxmusl-arm64": ["@img/sharp-linuxmusl-arm64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linuxmusl-arm64": "1.0.4" }, "os": "linux", "cpu": "arm64" }, "sha512-XrHMZwGQGvJg2V/oRSUfSAfjfPxO+4DkiRh6p2AFjLQztWUuY/o8Mq0eMQVIY7HJ1CDQUJlxGGZRw1a5bqmd1g=="],
+
+    "@img/sharp-linuxmusl-x64": ["@img/sharp-linuxmusl-x64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linuxmusl-x64": "1.0.4" }, "os": "linux", "cpu": "x64" }, "sha512-WT+d/cgqKkkKySYmqoZ8y3pxx7lx9vVejxW/W4DOFMYVSkErR+w7mf2u8m/y4+xHe7yY9DAXQMWQhpnMuFfScw=="],
+
+    "@img/sharp-win32-x64": ["@img/sharp-win32-x64@0.33.5", "", { "os": "win32", "cpu": "x64" }, "sha512-MpY/o8/8kj+EcnxwvrP4aTJSWw/aZ7JIGR4aBeZkZw5B7/Jn+tY9/VNwtcoGmdT7GfggGIU4kygOMSbYnOrAbg=="],
+
    "@types/bun": ["@types/bun@1.2.19", "", { "dependencies": { "bun-types": "1.2.19" } }, "sha512-d9ZCmrH3CJ2uYKXQIUuZ/pUnTqIvLDS0SK7pFmbx8ma+ziH/FRMoAq5bYpRG7y+w1gl+HgyNZbtqgMq4W4e2Lg=="],

    "@types/node": ["@types/node@20.19.9", "", { "dependencies": { "undici-types": "~6.21.0" } }, "sha512-cuVNgarYWZqxRJDQHEB58GEONhOK79QVR/qYx4S7kcUObQvUwvFnYxJuuHUKm2aieN9X3yZB4LZsuYNU1Qphsw=="],
@@ -50,5 +84,7 @@
    "undici": ["undici@5.29.0", "", { "dependencies": { "@fastify/busboy": "^2.0.0" } }, "sha512-raqeBD6NQK4SkWhQzeYKd1KmIG6dllBOTt55Rmkt4HtI9mwdWtJljnrXjAFUBLTSN67HWrOIZ3EPF4kjUw80Bg=="],

    "undici-types": ["undici-types@6.21.0", "", {}, "sha512-iwDZqg0QAGrg9Rav5H4n0M64c3mkR59cJ6wQp+7C4nI0gsmExaedaYLNO44eT4AtBBwjbTiGPMlt2Md0T9H9JQ=="],
+
+    "zod": ["zod@3.25.76", "", {}, "sha512-gzUt/qt81nXsFGKIFcC3YnfEAx5NkunCfnDlvuBSSFS02bcXu4Lmea0AFIUwbLWxWPx3d9p8S5QoaujKcNQxcQ=="],
  }
 }
--- a/base-action/package.json
+++ b/base-action/package.json
@@ -11,6 +11,7 @@
  },
  "dependencies": {
    "@actions/core": "^1.10.1",
+    "@anthropic-ai/claude-agent-sdk": "^0.1.70",
    "shell-quote": "^1.8.3"
  },
  "devDependencies": {
--- a/base-action/src/parse-sdk-options.ts
+++ b/base-action/src/parse-sdk-options.ts
@@ -0,0 +1,261 @@
+import { parse as parseShellArgs } from "shell-quote";
+import type { ClaudeOptions } from "./run-claude";
+import type { Options as SdkOptions } from "@anthropic-ai/claude-agent-sdk";
+
+/**
+ * Result of parsing ClaudeOptions for SDK usage
+ */
+export type ParsedSdkOptions = {
+  sdkOptions: SdkOptions;
+  showFullOutput: boolean;
+  hasJsonSchema: boolean;
+};
+
+// Flags that should accumulate multiple values instead of overwriting
+// Include both camelCase and hyphenated variants for CLI compatibility
+const ACCUMULATING_FLAGS = new Set([
+  "allowedTools",
+  "allowed-tools",
+  "disallowedTools",
+  "disallowed-tools",
+  "mcp-config",
+]);
+
+// Delimiter used to join accumulated flag values
+const ACCUMULATE_DELIMITER = "\x00";
+
+type McpConfig = {
+  mcpServers?: Record<string, unknown>;
+};
+
+/**
+ * Merge multiple MCP config values into a single config.
+ * Each config can be a JSON string or a file path.
+ * For JSON strings, mcpServers objects are merged.
+ * For file paths, they are kept as-is (user's file takes precedence and is used last).
+ */
+function mergeMcpConfigs(configValues: string[]): string {
+  const merged: McpConfig = { mcpServers: {} };
+  let lastFilePath: string | null = null;
+
+  for (const config of configValues) {
+    const trimmed = config.trim();
+    if (!trimmed) continue;
+
+    // Check if it's a JSON string (starts with {) or a file path
+    if (trimmed.startsWith("{")) {
+      try {
+        const parsed = JSON.parse(trimmed) as McpConfig;
+        if (parsed.mcpServers) {
+          Object.assign(merged.mcpServers!, parsed.mcpServers);
+        }
+      } catch {
+        // If JSON parsing fails, treat as file path
+        lastFilePath = trimmed;
+      }
+    } else {
+      // It's a file path - store it to handle separately
+      lastFilePath = trimmed;
+    }
+  }
+
+  // If we have file paths, we need to keep the merged JSON and let the file
+  // be handled separately. Since we can only return one value, merge what we can.
+  // If there's a file path, we need a different approach - read the file at runtime.
+  // For now, if there's a file path, we'll stringify the merged config.
+  // The action prepends its config as JSON, so we can safely merge inline JSON configs.
+
+  // If no inline configs were found (all file paths), return the last file path
+  if (Object.keys(merged.mcpServers!).length === 0 && lastFilePath) {
+    return lastFilePath;
+  }
+
+  // Note: If user passes a file path, we cannot merge it at parse time since
+  // we don't have access to the file system here. The action's built-in MCP
+  // servers are always passed as inline JSON, so they will be merged.
+  // If user also passes inline JSON, it will be merged.
+  // If user passes a file path, they should ensure it includes all needed servers.
+
+  return JSON.stringify(merged);
+}
+
+/**
+ * Parse claudeArgs string into extraArgs record for SDK pass-through
+ * The SDK/CLI will handle --mcp-config, --json-schema, etc.
+ * For allowedTools and disallowedTools, multiple occurrences are accumulated (null-char joined).
+ * Accumulating flags also consume all consecutive non-flag values
+ * (e.g., --allowed-tools "Tool1" "Tool2" "Tool3" captures all three).
+ */
+function parseClaudeArgsToExtraArgs(
+  claudeArgs?: string,
+): Record<string, string | null> {
+  if (!claudeArgs?.trim()) return {};
+
+  const result: Record<string, string | null> = {};
+  const args = parseShellArgs(claudeArgs).filter(
+    (arg): arg is string => typeof arg === "string",
+  );
+
+  for (let i = 0; i < args.length; i++) {
+    const arg = args[i];
+    if (arg?.startsWith("--")) {
+      const flag = arg.slice(2);
+      const nextArg = args[i + 1];
+
+      // Check if next arg is a value (not another flag)
+      if (nextArg && !nextArg.startsWith("--")) {
+        // For accumulating flags, consume all consecutive non-flag values
+        // This handles: --allowed-tools "Tool1" "Tool2" "Tool3"
+        if (ACCUMULATING_FLAGS.has(flag)) {
+          const values: string[] = [];
+          while (i + 1 < args.length && !args[i + 1]?.startsWith("--")) {
+            i++;
+            values.push(args[i]!);
+          }
+          const joinedValues = values.join(ACCUMULATE_DELIMITER);
+          if (result[flag]) {
+            result[flag] =
+              `${result[flag]}${ACCUMULATE_DELIMITER}${joinedValues}`;
+          } else {
+            result[flag] = joinedValues;
+          }
+        } else {
+          result[flag] = nextArg;
+          i++; // Skip the value
+        }
+      } else {
+        result[flag] = null; // Boolean flag
+      }
+    }
+  }
+
+  return result;
+}
+
+/**
+ * Parse ClaudeOptions into SDK-compatible options
+ * Uses extraArgs for CLI pass-through instead of duplicating option parsing
+ */
+export function parseSdkOptions(options: ClaudeOptions): ParsedSdkOptions {
+  // Determine output verbosity
+  const isDebugMode = process.env.ACTIONS_STEP_DEBUG === "true";
+  const showFullOutput = options.showFullOutput === "true" || isDebugMode;
+
+  // Parse claudeArgs into extraArgs for CLI pass-through
+  const extraArgs = parseClaudeArgsToExtraArgs(options.claudeArgs);
+
+  // Detect if --json-schema is present (for hasJsonSchema flag)
+  const hasJsonSchema = "json-schema" in extraArgs;
+
+  // Extract and merge allowedTools from all sources:
+  // 1. From extraArgs (parsed from claudeArgs - contains tag mode's tools)
+  //    - Check both camelCase (--allowedTools) and hyphenated (--allowed-tools) variants
+  // 2. From options.allowedTools (direct input - may be undefined)
+  // This prevents duplicate flags being overwritten when claudeArgs contains --allowedTools
+  const allowedToolsValues = [
+    extraArgs["allowedTools"],
+    extraArgs["allowed-tools"],
+  ]
+    .filter(Boolean)
+    .join(ACCUMULATE_DELIMITER);
+  const extraArgsAllowedTools = allowedToolsValues
+    ? allowedToolsValues
+        .split(ACCUMULATE_DELIMITER)
+        .flatMap((v) => v.split(","))
+        .map((t) => t.trim())
+        .filter(Boolean)
+    : [];
+  const directAllowedTools = options.allowedTools
+    ? options.allowedTools.split(",").map((t) => t.trim())
+    : [];
+  const mergedAllowedTools = [
+    ...new Set([...extraArgsAllowedTools, ...directAllowedTools]),
+  ];
+  delete extraArgs["allowedTools"];
+  delete extraArgs["allowed-tools"];
+
+  // Same for disallowedTools - check both camelCase and hyphenated variants
+  const disallowedToolsValues = [
+    extraArgs["disallowedTools"],
+    extraArgs["disallowed-tools"],
+  ]
+    .filter(Boolean)
+    .join(ACCUMULATE_DELIMITER);
+  const extraArgsDisallowedTools = disallowedToolsValues
+    ? disallowedToolsValues
+        .split(ACCUMULATE_DELIMITER)
+        .flatMap((v) => v.split(","))
+        .map((t) => t.trim())
+        .filter(Boolean)
+    : [];
+  const directDisallowedTools = options.disallowedTools
+    ? options.disallowedTools.split(",").map((t) => t.trim())
+    : [];
+  const mergedDisallowedTools = [
+    ...new Set([...extraArgsDisallowedTools, ...directDisallowedTools]),
+  ];
+  delete extraArgs["disallowedTools"];
+  delete extraArgs["disallowed-tools"];
+
+  // Merge multiple --mcp-config values by combining their mcpServers objects
+  // The action prepends its config (github_comment, github_ci, etc.) as inline JSON,
+  // and users may provide their own config as inline JSON or file path
+  if (extraArgs["mcp-config"]) {
+    const mcpConfigValues = extraArgs["mcp-config"].split(ACCUMULATE_DELIMITER);
+    if (mcpConfigValues.length > 1) {
+      extraArgs["mcp-config"] = mergeMcpConfigs(mcpConfigValues);
+    }
+  }
+
+  // Build custom environment
+  const env: Record<string, string | undefined> = { ...process.env };
+  if (process.env.INPUT_ACTION_INPUTS_PRESENT) {
+    env.GITHUB_ACTION_INPUTS = process.env.INPUT_ACTION_INPUTS_PRESENT;
+  }
+
+  // Build system prompt option - default to claude_code preset
+  let systemPrompt: SdkOptions["systemPrompt"];
+  if (options.systemPrompt) {
+    systemPrompt = options.systemPrompt;
+  } else if (options.appendSystemPrompt) {
+    systemPrompt = {
+      type: "preset",
+      preset: "claude_code",
+      append: options.appendSystemPrompt,
+    };
+  } else {
+    // Default to claude_code preset when no custom prompt is specified
+    systemPrompt = {
+      type: "preset",
+      preset: "claude_code",
+    };
+  }
+
+  // Build SDK options - use merged tools from both direct options and claudeArgs
+  const sdkOptions: SdkOptions = {
+    // Direct options from ClaudeOptions inputs
+    model: options.model,
+    maxTurns: options.maxTurns ? parseInt(options.maxTurns, 10) : undefined,
+    allowedTools:
+      mergedAllowedTools.length > 0 ? mergedAllowedTools : undefined,
+    disallowedTools:
+      mergedDisallowedTools.length > 0 ? mergedDisallowedTools : undefined,
+    systemPrompt,
+    fallbackModel: options.fallbackModel,
+    pathToClaudeCodeExecutable: options.pathToClaudeCodeExecutable,
+
+    // Pass through claudeArgs as extraArgs - CLI handles --mcp-config, --json-schema, etc.
+    // Note: allowedTools and disallowedTools have been removed from extraArgs to prevent duplicates
+    extraArgs,
+    env,
+
+    // Load settings from all sources to pick up CLI-installed plugins, CLAUDE.md, etc.
+    settingSources: ["user", "project", "local"],
+  };
+
+  return {
+    sdkOptions,
+    showFullOutput,
+    hasJsonSchema,
+  };
+}
--- a/base-action/src/run-claude-sdk.ts
+++ b/base-action/src/run-claude-sdk.ts
@@ -0,0 +1,151 @@
+import * as core from "@actions/core";
+import { readFile, writeFile } from "fs/promises";
+import { query } from "@anthropic-ai/claude-agent-sdk";
+import type {
+  SDKMessage,
+  SDKResultMessage,
+} from "@anthropic-ai/claude-agent-sdk";
+import type { ParsedSdkOptions } from "./parse-sdk-options";
+
+const EXECUTION_FILE = `${process.env.RUNNER_TEMP}/claude-execution-output.json`;
+
+/**
+ * Sanitizes SDK output to match CLI sanitization behavior
+ */
+function sanitizeSdkOutput(
+  message: SDKMessage,
+  showFullOutput: boolean,
+): string | null {
+  if (showFullOutput) {
+    return JSON.stringify(message, null, 2);
+  }
+
+  // System initialization - safe to show
+  if (message.type === "system" && message.subtype === "init") {
+    return JSON.stringify(
+      {
+        type: "system",
+        subtype: "init",
+        message: "Claude Code initialized",
+        model: "model" in message ? message.model : "unknown",
+      },
+      null,
+      2,
+    );
+  }
+
+  // Result messages - show sanitized summary
+  if (message.type === "result") {
+    const resultMsg = message as SDKResultMessage;
+    return JSON.stringify(
+      {
+        type: "result",
+        subtype: resultMsg.subtype,
+        is_error: resultMsg.is_error,
+        duration_ms: resultMsg.duration_ms,
+        num_turns: resultMsg.num_turns,
+        total_cost_usd: resultMsg.total_cost_usd,
+        permission_denials: resultMsg.permission_denials,
+      },
+      null,
+      2,
+    );
+  }
+
+  // Suppress other message types in non-full-output mode
+  return null;
+}
+
+/**
+ * Run Claude using the Agent SDK
+ */
+export async function runClaudeWithSdk(
+  promptPath: string,
+  { sdkOptions, showFullOutput, hasJsonSchema }: ParsedSdkOptions,
+): Promise<void> {
+  const prompt = await readFile(promptPath, "utf-8");
+
+  if (!showFullOutput) {
+    console.log(
+      "Running Claude Code via SDK (full output hidden for security)...",
+    );
+    console.log(
+      "Rerun in debug mode or enable `show_full_output: true` in your workflow file for full output.",
+    );
+  }
+
+  console.log(`Running Claude with prompt from file: ${promptPath}`);
+  // Log SDK options without env (which could contain sensitive data)
+  const { env, ...optionsToLog } = sdkOptions;
+  console.log("SDK options:", JSON.stringify(optionsToLog, null, 2));
+
+  const messages: SDKMessage[] = [];
+  let resultMessage: SDKResultMessage | undefined;
+
+  try {
+    for await (const message of query({ prompt, options: sdkOptions })) {
+      messages.push(message);
+
+      const sanitized = sanitizeSdkOutput(message, showFullOutput);
+      if (sanitized) {
+        console.log(sanitized);
+      }
+
+      if (message.type === "result") {
+        resultMessage = message as SDKResultMessage;
+      }
+    }
+  } catch (error) {
+    console.error("SDK execution error:", error);
+    core.setOutput("conclusion", "failure");
+    process.exit(1);
+  }
+
+  // Write execution file
+  try {
+    await writeFile(EXECUTION_FILE, JSON.stringify(messages, null, 2));
+    console.log(`Log saved to ${EXECUTION_FILE}`);
+    core.setOutput("execution_file", EXECUTION_FILE);
+  } catch (error) {
+    core.warning(`Failed to write execution file: ${error}`);
+  }
+
+  if (!resultMessage) {
+    core.setOutput("conclusion", "failure");
+    core.error("No result message received from Claude");
+    process.exit(1);
+  }
+
+  const isSuccess = resultMessage.subtype === "success";
+  core.setOutput("conclusion", isSuccess ? "success" : "failure");
+
+  // Handle structured output
+  if (hasJsonSchema) {
+    if (
+      isSuccess &&
+      "structured_output" in resultMessage &&
+      resultMessage.structured_output
+    ) {
+      const structuredOutputJson = JSON.stringify(
+        resultMessage.structured_output,
+      );
+      core.setOutput("structured_output", structuredOutputJson);
+      core.info(
+        `Set structured_output with ${Object.keys(resultMessage.structured_output as object).length} field(s)`,
+      );
+    } else {
+      core.setFailed(
+        `--json-schema was provided but Claude did not return structured_output. Result subtype: ${resultMessage.subtype}`,
+      );
+      core.setOutput("conclusion", "failure");
+      process.exit(1);
+    }
+  }
+
+  if (!isSuccess) {
+    if ("errors" in resultMessage && resultMessage.errors) {
+      core.error(`Execution failed: ${resultMessage.errors.join(", ")}`);
+    }
+    process.exit(1);
+  }
+}
--- a/base-action/src/run-claude.ts
+++ b/base-action/src/run-claude.ts
@@ -5,6 +5,8 @@ import { unlink, writeFile, stat, readFile } from "fs/promises";
 import { createWriteStream } from "fs";
 import { spawn } from "child_process";
 import { parse as parseShellArgs } from "shell-quote";
+import { runClaudeWithSdk } from "./run-claude-sdk";
+import { parseSdkOptions } from "./parse-sdk-options";

 const execAsync = promisify(exec);

@@ -122,6 +124,36 @@ export function prepareRunConfig(
  };
 }

+/**
+ * Parses session_id from execution file and sets GitHub Action output
+ * Exported for testing
+ */
+export async function parseAndSetSessionId(
+  executionFile: string,
+): Promise<void> {
+  try {
+    const content = await readFile(executionFile, "utf-8");
+    const messages = JSON.parse(content) as {
+      type: string;
+      subtype?: string;
+      session_id?: string;
+    }[];
+
+    // Find the system.init message which contains session_id
+    const initMessage = messages.find(
+      (m) => m.type === "system" && m.subtype === "init",
+    );
+
+    if (initMessage?.session_id) {
+      core.setOutput("session_id", initMessage.session_id);
+      core.info(`Set session_id: ${initMessage.session_id}`);
+    }
+  } catch (error) {
+    // Don't fail the action if session_id extraction fails
+    core.warning(`Failed to extract session_id: ${error}`);
+  }
+}
+
 /**
 * Parses structured_output from execution file and sets GitHub Action outputs
 * Only runs if --json-schema was explicitly provided in claude_args
@@ -165,6 +197,17 @@ export async function parseAndSetStructuredOutputs(
 }

 export async function runClaude(promptPath: string, options: ClaudeOptions) {
+  // Feature flag: use SDK path by default, set USE_AGENT_SDK=false to use CLI
+  const useAgentSdk = process.env.USE_AGENT_SDK !== "false";
+  console.log(
+    `Using ${useAgentSdk ? "Agent SDK" : "CLI"} path (USE_AGENT_SDK=${process.env.USE_AGENT_SDK ?? "unset"})`,
+  );
+
+  if (useAgentSdk) {
+    const parsedOptions = parseSdkOptions(options);
+    return runClaudeWithSdk(promptPath, parsedOptions);
+  }
+
  const config = prepareRunConfig(promptPath, options);

  // Detect if --json-schema is present in claude args
@@ -355,6 +398,9 @@ export async function runClaude(promptPath: string, options: ClaudeOptions) {

    core.setOutput("execution_file", EXECUTION_FILE);

+    // Extract and set session_id
+    await parseAndSetSessionId(EXECUTION_FILE);
+
    // Parse and set structured outputs only if user provided --json-schema in claude_args
    if (hasJsonSchema) {
      try {
--- a/base-action/test/parse-sdk-options.test.ts
+++ b/base-action/test/parse-sdk-options.test.ts
@@ -0,0 +1,315 @@
+#!/usr/bin/env bun
+
+import { describe, test, expect } from "bun:test";
+import { parseSdkOptions } from "../src/parse-sdk-options";
+import type { ClaudeOptions } from "../src/run-claude";
+
+describe("parseSdkOptions", () => {
+  describe("allowedTools merging", () => {
+    test("should extract allowedTools from claudeArgs", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--allowedTools "Edit,Read,Write"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toEqual(["Edit", "Read", "Write"]);
+      expect(result.sdkOptions.extraArgs?.["allowedTools"]).toBeUndefined();
+    });
+
+    test("should extract allowedTools from claudeArgs with MCP tools", () => {
+      const options: ClaudeOptions = {
+        claudeArgs:
+          '--allowedTools "Edit,Read,mcp__github_comment__update_claude_comment"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toEqual([
+        "Edit",
+        "Read",
+        "mcp__github_comment__update_claude_comment",
+      ]);
+    });
+
+    test("should accumulate multiple --allowedTools flags from claudeArgs", () => {
+      // This simulates tag mode adding its tools, then user adding their own
+      const options: ClaudeOptions = {
+        claudeArgs:
+          '--allowedTools "Edit,Read,mcp__github_comment__update_claude_comment" --model "claude-3" --allowedTools "Bash(npm install),mcp__github__get_issue"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toEqual([
+        "Edit",
+        "Read",
+        "mcp__github_comment__update_claude_comment",
+        "Bash(npm install)",
+        "mcp__github__get_issue",
+      ]);
+    });
+
+    test("should merge allowedTools from both claudeArgs and direct options", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--allowedTools "Edit,Read"',
+        allowedTools: "Write,Glob",
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toEqual([
+        "Edit",
+        "Read",
+        "Write",
+        "Glob",
+      ]);
+    });
+
+    test("should deduplicate allowedTools when merging", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--allowedTools "Edit,Read"',
+        allowedTools: "Edit,Write",
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toEqual(["Edit", "Read", "Write"]);
+    });
+
+    test("should use only direct options when claudeArgs has no allowedTools", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--model "claude-3-5-sonnet"',
+        allowedTools: "Edit,Read",
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toEqual(["Edit", "Read"]);
+    });
+
+    test("should return undefined allowedTools when neither source has it", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--model "claude-3-5-sonnet"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toBeUndefined();
+    });
+
+    test("should remove allowedTools from extraArgs after extraction", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--allowedTools "Edit,Read" --model "claude-3-5-sonnet"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.extraArgs?.["allowedTools"]).toBeUndefined();
+      expect(result.sdkOptions.extraArgs?.["model"]).toBe("claude-3-5-sonnet");
+    });
+
+    test("should handle hyphenated --allowed-tools flag", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--allowed-tools "Edit,Read,Write"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toEqual(["Edit", "Read", "Write"]);
+      expect(result.sdkOptions.extraArgs?.["allowed-tools"]).toBeUndefined();
+    });
+
+    test("should accumulate multiple --allowed-tools flags (hyphenated)", () => {
+      // This is the exact scenario from issue #746
+      const options: ClaudeOptions = {
+        claudeArgs:
+          '--allowed-tools "Bash(git log:*)" "Bash(git diff:*)" "Bash(git fetch:*)" "Bash(gh pr:*)"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.allowedTools).toEqual([
+        "Bash(git log:*)",
+        "Bash(git diff:*)",
+        "Bash(git fetch:*)",
+        "Bash(gh pr:*)",
+      ]);
+    });
+
+    test("should handle mixed camelCase and hyphenated allowedTools flags", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--allowedTools "Edit,Read" --allowed-tools "Write,Glob"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      // Both should be merged - note: order depends on which key is found first
+      expect(result.sdkOptions.allowedTools).toContain("Edit");
+      expect(result.sdkOptions.allowedTools).toContain("Read");
+      expect(result.sdkOptions.allowedTools).toContain("Write");
+      expect(result.sdkOptions.allowedTools).toContain("Glob");
+    });
+  });
+
+  describe("disallowedTools merging", () => {
+    test("should extract disallowedTools from claudeArgs", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--disallowedTools "Bash,Write"',
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.disallowedTools).toEqual(["Bash", "Write"]);
+      expect(result.sdkOptions.extraArgs?.["disallowedTools"]).toBeUndefined();
+    });
+
+    test("should merge disallowedTools from both sources", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: '--disallowedTools "Bash"',
+        disallowedTools: "Write",
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.disallowedTools).toEqual(["Bash", "Write"]);
+    });
+  });
+
+  describe("mcp-config merging", () => {
+    test("should pass through single mcp-config in extraArgs", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: `--mcp-config '{"mcpServers":{"server1":{"command":"cmd1"}}}'`,
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.extraArgs?.["mcp-config"]).toBe(
+        '{"mcpServers":{"server1":{"command":"cmd1"}}}',
+      );
+    });
+
+    test("should merge multiple mcp-config flags with inline JSON", () => {
+      // Simulates action prepending its config, then user providing their own
+      const options: ClaudeOptions = {
+        claudeArgs: `--mcp-config '{"mcpServers":{"github_comment":{"command":"node","args":["server.js"]}}}' --mcp-config '{"mcpServers":{"user_server":{"command":"custom","args":["run"]}}}'`,
+      };
+
+      const result = parseSdkOptions(options);
+
+      const mcpConfig = JSON.parse(
+        result.sdkOptions.extraArgs?.["mcp-config"] as string,
+      );
+      expect(mcpConfig.mcpServers).toHaveProperty("github_comment");
+      expect(mcpConfig.mcpServers).toHaveProperty("user_server");
+      expect(mcpConfig.mcpServers.github_comment.command).toBe("node");
+      expect(mcpConfig.mcpServers.user_server.command).toBe("custom");
+    });
+
+    test("should merge three mcp-config flags", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: `--mcp-config '{"mcpServers":{"server1":{"command":"cmd1"}}}' --mcp-config '{"mcpServers":{"server2":{"command":"cmd2"}}}' --mcp-config '{"mcpServers":{"server3":{"command":"cmd3"}}}'`,
+      };
+
+      const result = parseSdkOptions(options);
+
+      const mcpConfig = JSON.parse(
+        result.sdkOptions.extraArgs?.["mcp-config"] as string,
+      );
+      expect(mcpConfig.mcpServers).toHaveProperty("server1");
+      expect(mcpConfig.mcpServers).toHaveProperty("server2");
+      expect(mcpConfig.mcpServers).toHaveProperty("server3");
+    });
+
+    test("should handle mcp-config file path when no inline JSON exists", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: `--mcp-config /tmp/user-mcp-config.json`,
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.extraArgs?.["mcp-config"]).toBe(
+        "/tmp/user-mcp-config.json",
+      );
+    });
+
+    test("should merge inline JSON configs when file path is also present", () => {
+      // When action provides inline JSON and user provides a file path,
+      // the inline JSON configs should be merged (file paths cannot be merged at parse time)
+      const options: ClaudeOptions = {
+        claudeArgs: `--mcp-config '{"mcpServers":{"github_comment":{"command":"node"}}}' --mcp-config '{"mcpServers":{"github_ci":{"command":"node"}}}' --mcp-config /tmp/user-config.json`,
+      };
+
+      const result = parseSdkOptions(options);
+
+      // The inline JSON configs should be merged
+      const mcpConfig = JSON.parse(
+        result.sdkOptions.extraArgs?.["mcp-config"] as string,
+      );
+      expect(mcpConfig.mcpServers).toHaveProperty("github_comment");
+      expect(mcpConfig.mcpServers).toHaveProperty("github_ci");
+    });
+
+    test("should handle mcp-config with other flags", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: `--mcp-config '{"mcpServers":{"server1":{}}}' --model claude-3-5-sonnet --mcp-config '{"mcpServers":{"server2":{}}}'`,
+      };
+
+      const result = parseSdkOptions(options);
+
+      const mcpConfig = JSON.parse(
+        result.sdkOptions.extraArgs?.["mcp-config"] as string,
+      );
+      expect(mcpConfig.mcpServers).toHaveProperty("server1");
+      expect(mcpConfig.mcpServers).toHaveProperty("server2");
+      expect(result.sdkOptions.extraArgs?.["model"]).toBe("claude-3-5-sonnet");
+    });
+
+    test("should handle real-world scenario: action config + user config", () => {
+      // This is the exact scenario from the bug report
+      const actionConfig = JSON.stringify({
+        mcpServers: {
+          github_comment: {
+            command: "node",
+            args: ["github-comment-server.js"],
+          },
+          github_ci: { command: "node", args: ["github-ci-server.js"] },
+        },
+      });
+      const userConfig = JSON.stringify({
+        mcpServers: {
+          my_custom_server: { command: "python", args: ["server.py"] },
+        },
+      });
+
+      const options: ClaudeOptions = {
+        claudeArgs: `--mcp-config '${actionConfig}' --mcp-config '${userConfig}'`,
+      };
+
+      const result = parseSdkOptions(options);
+
+      const mcpConfig = JSON.parse(
+        result.sdkOptions.extraArgs?.["mcp-config"] as string,
+      );
+      // All servers should be present
+      expect(mcpConfig.mcpServers).toHaveProperty("github_comment");
+      expect(mcpConfig.mcpServers).toHaveProperty("github_ci");
+      expect(mcpConfig.mcpServers).toHaveProperty("my_custom_server");
+    });
+  });
+
+  describe("other extraArgs passthrough", () => {
+    test("should pass through json-schema in extraArgs", () => {
+      const options: ClaudeOptions = {
+        claudeArgs: `--json-schema '{"type":"object"}'`,
+      };
+
+      const result = parseSdkOptions(options);
+
+      expect(result.sdkOptions.extraArgs?.["json-schema"]).toBe(
+        '{"type":"object"}',
+      );
+      expect(result.hasJsonSchema).toBe(true);
+    });
+  });
+});
--- a/base-action/test/structured-output.test.ts
+++ b/base-action/test/structured-output.test.ts
@@ -4,7 +4,10 @@ import { describe, test, expect, afterEach, beforeEach, spyOn } from "bun:test";
 import { writeFile, unlink } from "fs/promises";
 import { tmpdir } from "os";
 import { join } from "path";
-import { parseAndSetStructuredOutputs } from "../src/run-claude";
+import {
+  parseAndSetStructuredOutputs,
+  parseAndSetSessionId,
+} from "../src/run-claude";
 import * as core from "@actions/core";

 // Mock execution file path
@@ -35,16 +38,19 @@ async function createMockExecutionFile(
 // Spy on core functions
 let setOutputSpy: any;
 let infoSpy: any;
+let warningSpy: any;

 beforeEach(() => {
  setOutputSpy = spyOn(core, "setOutput").mockImplementation(() => {});
  infoSpy = spyOn(core, "info").mockImplementation(() => {});
+  warningSpy = spyOn(core, "warning").mockImplementation(() => {});
 });

 describe("parseAndSetStructuredOutputs", () => {
  afterEach(async () => {
    setOutputSpy?.mockRestore();
    infoSpy?.mockRestore();
+    warningSpy?.mockRestore();
    try {
      await unlink(TEST_EXECUTION_FILE);
    } catch {
@@ -156,3 +162,66 @@ describe("parseAndSetStructuredOutputs", () => {
    );
  });
 });
+
+describe("parseAndSetSessionId", () => {
+  afterEach(async () => {
+    setOutputSpy?.mockRestore();
+    infoSpy?.mockRestore();
+    warningSpy?.mockRestore();
+    try {
+      await unlink(TEST_EXECUTION_FILE);
+    } catch {
+      // Ignore if file doesn't exist
+    }
+  });
+
+  test("should extract session_id from system.init message", async () => {
+    const messages = [
+      { type: "system", subtype: "init", session_id: "test-session-123" },
+      { type: "result", cost_usd: 0.01 },
+    ];
+    await writeFile(TEST_EXECUTION_FILE, JSON.stringify(messages));
+
+    await parseAndSetSessionId(TEST_EXECUTION_FILE);
+
+    expect(setOutputSpy).toHaveBeenCalledWith("session_id", "test-session-123");
+    expect(infoSpy).toHaveBeenCalledWith("Set session_id: test-session-123");
+  });
+
+  test("should handle missing session_id gracefully", async () => {
+    const messages = [
+      { type: "system", subtype: "init" },
+      { type: "result", cost_usd: 0.01 },
+    ];
+    await writeFile(TEST_EXECUTION_FILE, JSON.stringify(messages));
+
+    await parseAndSetSessionId(TEST_EXECUTION_FILE);
+
+    expect(setOutputSpy).not.toHaveBeenCalled();
+  });
+
+  test("should handle missing system.init message gracefully", async () => {
+    const messages = [{ type: "result", cost_usd: 0.01 }];
+    await writeFile(TEST_EXECUTION_FILE, JSON.stringify(messages));
+
+    await parseAndSetSessionId(TEST_EXECUTION_FILE);
+
+    expect(setOutputSpy).not.toHaveBeenCalled();
+  });
+
+  test("should handle malformed JSON gracefully with warning", async () => {
+    await writeFile(TEST_EXECUTION_FILE, "{ invalid json");
+
+    await parseAndSetSessionId(TEST_EXECUTION_FILE);
+
+    expect(setOutputSpy).not.toHaveBeenCalled();
+    expect(warningSpy).toHaveBeenCalled();
+  });
+
+  test("should handle non-existent file gracefully with warning", async () => {
+    await parseAndSetSessionId("/nonexistent/file.json");
+
+    expect(setOutputSpy).not.toHaveBeenCalled();
+    expect(warningSpy).toHaveBeenCalled();
+  });
+});
--- a/bun.lock
+++ b/bun.lock
@@ -1,11 +1,13 @@
 {
  "lockfileVersion": 1,
+  "configVersion": 0,
  "workspaces": {
    "": {
      "name": "@anthropic-ai/claude-code-action",
      "dependencies": {
        "@actions/core": "^1.10.1",
        "@actions/github": "^6.0.1",
+        "@anthropic-ai/claude-agent-sdk": "^0.1.70",
        "@modelcontextprotocol/sdk": "^1.11.0",
        "@octokit/graphql": "^8.2.2",
        "@octokit/rest": "^21.1.1",
@@ -35,8 +37,40 @@

    "@actions/io": ["@actions/io@1.1.3", "", {}, "sha512-wi9JjgKLYS7U/z8PPbco+PvTb/nRWjeoFlJ1Qer83k/3C5PHQi28hiVdeE2kHXmIL99mQFawx8qt/JPjZilJ8Q=="],

+    "@anthropic-ai/claude-agent-sdk": ["@anthropic-ai/claude-agent-sdk@0.1.70", "", { "optionalDependencies": { "@img/sharp-darwin-arm64": "^0.33.5", "@img/sharp-darwin-x64": "^0.33.5", "@img/sharp-linux-arm": "^0.33.5", "@img/sharp-linux-arm64": "^0.33.5", "@img/sharp-linux-x64": "^0.33.5", "@img/sharp-linuxmusl-arm64": "^0.33.5", "@img/sharp-linuxmusl-x64": "^0.33.5", "@img/sharp-win32-x64": "^0.33.5" }, "peerDependencies": { "zod": "^3.24.1" } }, "sha512-4jpFPDX8asys6skO1r3Pzh0Fe9nbND2ASYTWuyFB5iN9bWEL6WScTFyGokjql3M2TkEp9ZGuB2YYpTCdaqT9Sw=="],
+
    "@fastify/busboy": ["@fastify/busboy@2.1.1", "", {}, "sha512-vBZP4NlzfOlerQTnba4aqZoMhE/a9HY7HRqoOPaETQcSQuWEIyZMHGfVu6w9wGtGK5fED5qRs2DteVCjOH60sA=="],

+    "@img/sharp-darwin-arm64": ["@img/sharp-darwin-arm64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-darwin-arm64": "1.0.4" }, "os": "darwin", "cpu": "arm64" }, "sha512-UT4p+iz/2H4twwAoLCqfA9UH5pI6DggwKEGuaPy7nCVQ8ZsiY5PIcrRvD1DzuY3qYL07NtIQcWnBSY/heikIFQ=="],
+
+    "@img/sharp-darwin-x64": ["@img/sharp-darwin-x64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-darwin-x64": "1.0.4" }, "os": "darwin", "cpu": "x64" }, "sha512-fyHac4jIc1ANYGRDxtiqelIbdWkIuQaI84Mv45KvGRRxSAa7o7d1ZKAOBaYbnepLC1WqxfpimdeWfvqqSGwR2Q=="],
+
+    "@img/sharp-libvips-darwin-arm64": ["@img/sharp-libvips-darwin-arm64@1.0.4", "", { "os": "darwin", "cpu": "arm64" }, "sha512-XblONe153h0O2zuFfTAbQYAX2JhYmDHeWikp1LM9Hul9gVPjFY427k6dFEcOL72O01QxQsWi761svJ/ev9xEDg=="],
+
+    "@img/sharp-libvips-darwin-x64": ["@img/sharp-libvips-darwin-x64@1.0.4", "", { "os": "darwin", "cpu": "x64" }, "sha512-xnGR8YuZYfJGmWPvmlunFaWJsb9T/AO2ykoP3Fz/0X5XV2aoYBPkX6xqCQvUTKKiLddarLaxpzNe+b1hjeWHAQ=="],
+
+    "@img/sharp-libvips-linux-arm": ["@img/sharp-libvips-linux-arm@1.0.5", "", { "os": "linux", "cpu": "arm" }, "sha512-gvcC4ACAOPRNATg/ov8/MnbxFDJqf/pDePbBnuBDcjsI8PssmjoKMAz4LtLaVi+OnSb5FK/yIOamqDwGmXW32g=="],
+
+    "@img/sharp-libvips-linux-arm64": ["@img/sharp-libvips-linux-arm64@1.0.4", "", { "os": "linux", "cpu": "arm64" }, "sha512-9B+taZ8DlyyqzZQnoeIvDVR/2F4EbMepXMc/NdVbkzsJbzkUjhXv/70GQJ7tdLA4YJgNP25zukcxpX2/SueNrA=="],
+
+    "@img/sharp-libvips-linux-x64": ["@img/sharp-libvips-linux-x64@1.0.4", "", { "os": "linux", "cpu": "x64" }, "sha512-MmWmQ3iPFZr0Iev+BAgVMb3ZyC4KeFc3jFxnNbEPas60e1cIfevbtuyf9nDGIzOaW9PdnDciJm+wFFaTlj5xYw=="],
+
+    "@img/sharp-libvips-linuxmusl-arm64": ["@img/sharp-libvips-linuxmusl-arm64@1.0.4", "", { "os": "linux", "cpu": "arm64" }, "sha512-9Ti+BbTYDcsbp4wfYib8Ctm1ilkugkA/uscUn6UXK1ldpC1JjiXbLfFZtRlBhjPZ5o1NCLiDbg8fhUPKStHoTA=="],
+
+    "@img/sharp-libvips-linuxmusl-x64": ["@img/sharp-libvips-linuxmusl-x64@1.0.4", "", { "os": "linux", "cpu": "x64" }, "sha512-viYN1KX9m+/hGkJtvYYp+CCLgnJXwiQB39damAO7WMdKWlIhmYTfHjwSbQeUK/20vY154mwezd9HflVFM1wVSw=="],
+
+    "@img/sharp-linux-arm": ["@img/sharp-linux-arm@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linux-arm": "1.0.5" }, "os": "linux", "cpu": "arm" }, "sha512-JTS1eldqZbJxjvKaAkxhZmBqPRGmxgu+qFKSInv8moZ2AmT5Yib3EQ1c6gp493HvrvV8QgdOXdyaIBrhvFhBMQ=="],
+
+    "@img/sharp-linux-arm64": ["@img/sharp-linux-arm64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linux-arm64": "1.0.4" }, "os": "linux", "cpu": "arm64" }, "sha512-JMVv+AMRyGOHtO1RFBiJy/MBsgz0x4AWrT6QoEVVTyh1E39TrCUpTRI7mx9VksGX4awWASxqCYLCV4wBZHAYxA=="],
+
+    "@img/sharp-linux-x64": ["@img/sharp-linux-x64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linux-x64": "1.0.4" }, "os": "linux", "cpu": "x64" }, "sha512-opC+Ok5pRNAzuvq1AG0ar+1owsu842/Ab+4qvU879ippJBHvyY5n2mxF1izXqkPYlGuP/M556uh53jRLJmzTWA=="],
+
+    "@img/sharp-linuxmusl-arm64": ["@img/sharp-linuxmusl-arm64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linuxmusl-arm64": "1.0.4" }, "os": "linux", "cpu": "arm64" }, "sha512-XrHMZwGQGvJg2V/oRSUfSAfjfPxO+4DkiRh6p2AFjLQztWUuY/o8Mq0eMQVIY7HJ1CDQUJlxGGZRw1a5bqmd1g=="],
+
+    "@img/sharp-linuxmusl-x64": ["@img/sharp-linuxmusl-x64@0.33.5", "", { "optionalDependencies": { "@img/sharp-libvips-linuxmusl-x64": "1.0.4" }, "os": "linux", "cpu": "x64" }, "sha512-WT+d/cgqKkkKySYmqoZ8y3pxx7lx9vVejxW/W4DOFMYVSkErR+w7mf2u8m/y4+xHe7yY9DAXQMWQhpnMuFfScw=="],
+
+    "@img/sharp-win32-x64": ["@img/sharp-win32-x64@0.33.5", "", { "os": "win32", "cpu": "x64" }, "sha512-MpY/o8/8kj+EcnxwvrP4aTJSWw/aZ7JIGR4aBeZkZw5B7/Jn+tY9/VNwtcoGmdT7GfggGIU4kygOMSbYnOrAbg=="],
+
    "@modelcontextprotocol/sdk": ["@modelcontextprotocol/sdk@1.16.0", "", { "dependencies": { "ajv": "^6.12.6", "content-type": "^1.0.5", "cors": "^2.8.5", "cross-spawn": "^7.0.5", "eventsource": "^3.0.2", "eventsource-parser": "^3.0.0", "express": "^5.0.1", "express-rate-limit": "^7.5.0", "pkce-challenge": "^5.0.0", "raw-body": "^3.0.0", "zod": "^3.23.8", "zod-to-json-schema": "^3.24.1" } }, "sha512-8ofX7gkZcLj9H9rSd50mCgm3SSF8C7XoclxJuLoV0Cz3rEQ1tv9MZRYYvJtm9n1BiEQQMzSmE/w2AEkNacLYfg=="],

    "@octokit/auth-token": ["@octokit/auth-token@4.0.0", "", {}, "sha512-tY/msAuJo6ARbK6SPIxZrPBms3xPbfwBrulZe0Wtr/DIY9lje2HeV1uoebShn6mx7SjCHif6EjMvoREj+gZ+SA=="],
--- a/package.json
+++ b/package.json
@@ -12,6 +12,7 @@
  "dependencies": {
    "@actions/core": "^1.10.1",
    "@actions/github": "^6.0.1",
+    "@anthropic-ai/claude-agent-sdk": "^0.1.70",
    "@modelcontextprotocol/sdk": "^1.11.0",
    "@octokit/graphql": "^8.2.2",
    "@octokit/rest": "^21.1.1",
--- a/src/create-prompt/index.ts
+++ b/src/create-prompt/index.ts
@@ -192,11 +192,6 @@ export function prepareContext(
      if (!isPR) {
        throw new Error("IS_PR must be true for pull_request_review event");
      }
-      if (!commentBody) {
-        throw new Error(
-          "COMMENT_BODY is required for pull_request_review event",
-        );
-      }
      eventData = {
        eventName: "pull_request_review",
        isPR: true,
@@ -464,6 +459,123 @@ export function generatePrompt(
  return mode.generatePrompt(context, githubData, useCommitSigning);
 }

+/**
+ * Generates a simplified prompt for tag mode (opt-in via USE_SIMPLE_PROMPT env var)
+ * @internal
+ */
+function generateSimplePrompt(
+  context: PreparedContext,
+  githubData: FetchDataResult,
+  useCommitSigning: boolean = false,
+): string {
+  const {
+    contextData,
+    comments,
+    changedFilesWithSHA,
+    reviewData,
+    imageUrlMap,
+  } = githubData;
+  const { eventData } = context;
+
+  const { triggerContext } = getEventTypeAndContext(context);
+
+  const formattedContext = formatContext(contextData, eventData.isPR);
+  const formattedComments = formatComments(comments, imageUrlMap);
+  const formattedReviewComments = eventData.isPR
+    ? formatReviewComments(reviewData, imageUrlMap)
+    : "";
+  const formattedChangedFiles = eventData.isPR
+    ? formatChangedFilesWithSHA(changedFilesWithSHA)
+    : "";
+
+  const hasImages = imageUrlMap && imageUrlMap.size > 0;
+  const imagesInfo = hasImages
+    ? `\n\n<images_info>
+Images from comments have been saved to disk. Paths are in the formatted content above. Use Read tool to view them.
+</images_info>`
+    : "";
+
+  const formattedBody = contextData?.body
+    ? formatBody(contextData.body, imageUrlMap)
+    : "No description provided";
+
+  const entityType = eventData.isPR ? "pull request" : "issue";
+  const jobUrl = `${GITHUB_SERVER_URL}/${context.repository}/actions/runs/${process.env.GITHUB_RUN_ID}`;
+
+  let promptContent = `You were tagged on a GitHub ${entityType} via "${context.triggerPhrase}". Read the request and decide how to help.
+
+<context>
+${formattedContext}
+</context>
+
+<${eventData.isPR ? "pr" : "issue"}_body>
+${formattedBody}
+</${eventData.isPR ? "pr" : "issue"}_body>
+
+<comments>
+${formattedComments || "No comments"}
+</comments>
+${
+  eventData.isPR
+    ? `
+<review_comments>
+${formattedReviewComments || "No review comments"}
+</review_comments>
+
+<changed_files>
+${formattedChangedFiles || "No files changed"}
+</changed_files>`
+    : ""
+}${imagesInfo}
+
+<metadata>
+repository: ${context.repository}
+${eventData.isPR && eventData.prNumber ? `pr_number: ${eventData.prNumber}` : ""}
+${!eventData.isPR && eventData.issueNumber ? `issue_number: ${eventData.issueNumber}` : ""}
+trigger: ${triggerContext}
+triggered_by: ${context.triggerUsername ?? "Unknown"}
+claude_comment_id: ${context.claudeCommentId}
+</metadata>
+${
+  (eventData.eventName === "issue_comment" ||
+    eventData.eventName === "pull_request_review_comment" ||
+    eventData.eventName === "pull_request_review") &&
+  eventData.commentBody
+    ? `
+<trigger_comment>
+${sanitizeContent(eventData.commentBody)}
+</trigger_comment>`
+    : ""
+}
+
+Your request is in <trigger_comment> above${eventData.eventName === "issues" ? ` (or the ${entityType} body for assigned/labeled events)` : ""}.
+
+Decide what's being asked:
+1. **Question or code review** - Answer directly or provide feedback
+2. **Code change** - Implement the change, commit, and push
+
+Communication:
+- Your ONLY visible output is your GitHub comment - update it with progress and results
+- Use mcp__github_comment__update_claude_comment to update (only "body" param needed)
+- Use checklist format for tasks: - [ ] incomplete, - [x] complete
+- Use ### headers (not #)
+${getCommitInstructions(eventData, githubData, context, useCommitSigning)}
+${
+  eventData.claudeBranch
+    ? `
+When done with changes, provide a PR link:
+[Create a PR](${GITHUB_SERVER_URL}/${context.repository}/compare/${eventData.baseBranch}...${eventData.claudeBranch}?quick_pull=1&title=<url-encoded-title>&body=<url-encoded-body>)
+Use THREE dots (...) between branches. URL-encode all parameters.`
+    : ""
+}
+
+Always include at the bottom:
+- Job link: [View job run](${jobUrl})
+- Follow the repo's CLAUDE.md file for project-specific guidelines`;
+
+  return promptContent;
+}
+
 /**
 * Generates the default prompt for tag mode
 * @internal
@@ -473,6 +585,10 @@ export function generateDefaultPrompt(
  githubData: FetchDataResult,
  useCommitSigning: boolean = false,
 ): string {
+  // Use simplified prompt if opted in
+  if (process.env.USE_SIMPLE_PROMPT === "true") {
+    return generateSimplePrompt(context, githubData, useCommitSigning);
+  }
  const {
    contextData,
    comments,
--- a/src/create-prompt/types.ts
+++ b/src/create-prompt/types.ts
@@ -23,7 +23,7 @@ type PullRequestReviewEvent = {
  eventName: "pull_request_review";
  isPR: true;
  prNumber: string;
-  commentBody: string;
+  commentBody?: string; // May be absent for approvals without comments
  claudeBranch?: string;
  baseBranch?: string;
 };
--- a/src/entrypoints/update-comment-link.ts
+++ b/src/entrypoints/update-comment-link.ts
@@ -152,7 +152,7 @@ async function run() {

    // Check if action failed and read output file for execution details
    let executionDetails: {
-      cost_usd?: number;
+      total_cost_usd?: number;
      duration_ms?: number;
      duration_api_ms?: number;
    } | null = null;
@@ -179,11 +179,11 @@ async function run() {
            const lastElement = outputData[outputData.length - 1];
            if (
              lastElement.type === "result" &&
-              "cost_usd" in lastElement &&
+              "total_cost_usd" in lastElement &&
              "duration_ms" in lastElement
            ) {
              executionDetails = {
-                cost_usd: lastElement.cost_usd,
+                total_cost_usd: lastElement.total_cost_usd,
                duration_ms: lastElement.duration_ms,
                duration_api_ms: lastElement.duration_api_ms,
              };
--- a/src/github/api/queries/github.ts
+++ b/src/github/api/queries/github.ts
@@ -13,6 +13,8 @@ export const PR_QUERY = `
        headRefName
        headRefOid
        createdAt
+        updatedAt
+        lastEditedAt
        additions
        deletions
        state
@@ -96,6 +98,8 @@ export const ISSUE_QUERY = `
          login
        }
        createdAt
+        updatedAt
+        lastEditedAt
        state
        comments(first: 100) {
          nodes {
--- a/src/github/data/fetcher.ts
+++ b/src/github/data/fetcher.ts
@@ -107,6 +107,38 @@ export function filterReviewsToTriggerTime<
  });
 }

+/**
+ * Checks if the issue/PR body was edited after the trigger time.
+ * This prevents a race condition where an attacker could edit the issue/PR body
+ * between when an authorized user triggered Claude and when Claude processes the request.
+ *
+ * @param contextData - The PR or issue data containing body and edit timestamps
+ * @param triggerTime - ISO timestamp of when the trigger event occurred
+ * @returns true if the body is safe to use, false if it was edited after trigger
+ */
+export function isBodySafeToUse(
+  contextData: { createdAt: string; updatedAt?: string; lastEditedAt?: string },
+  triggerTime: string | undefined,
+): boolean {
+  // If no trigger time is available, we can't validate - allow the body
+  // This maintains backwards compatibility for triggers that don't have timestamps
+  if (!triggerTime) return true;
+
+  const triggerTimestamp = new Date(triggerTime).getTime();
+
+  // Check if the body was edited after the trigger
+  // Use lastEditedAt if available (more accurate for body edits), otherwise fall back to updatedAt
+  const lastEditTime = contextData.lastEditedAt || contextData.updatedAt;
+  if (lastEditTime) {
+    const lastEditTimestamp = new Date(lastEditTime).getTime();
+    if (lastEditTimestamp >= triggerTimestamp) {
+      return false;
+    }
+  }
+
+  return true;
+}
+
 type FetchDataParams = {
  octokits: Octokits;
  repository: string;
@@ -273,9 +305,13 @@ export async function fetchGitHubData({
      body: c.body,
    }));

-  // Add the main issue/PR body if it has content
-  const mainBody: CommentWithImages[] = contextData.body
-    ? [
+  // Add the main issue/PR body if it has content and wasn't edited after trigger
+  // This prevents a TOCTOU race condition where an attacker could edit the body
+  // between when an authorized user triggered Claude and when Claude processes the request
+  let mainBody: CommentWithImages[] = [];
+  if (contextData.body) {
+    if (isBodySafeToUse(contextData, triggerTime)) {
+      mainBody = [
        {
          ...(isPR
            ? {
@@ -289,8 +325,14 @@ export async function fetchGitHubData({
                body: contextData.body,
              }),
        },
-      ]
-    : [];
+      ];
+    } else {
+      console.warn(
+        `Security: ${isPR ? "PR" : "Issue"} #${prNumber} body was edited after the trigger event. ` +
+          `Excluding body content to prevent potential injection attacks.`,
+      );
+    }
+  }

  const allComments = [
    ...mainBody,
--- a/src/github/operations/branch.ts
+++ b/src/github/operations/branch.ts
@@ -6,13 +6,112 @@
 * - For Issues: Create a new branch
 */

-import { $ } from "bun";
+import { execFileSync } from "child_process";
 import * as core from "@actions/core";
 import type { ParsedGitHubContext } from "../context";
 import type { GitHubPullRequest } from "../types";
 import type { Octokits } from "../api/client";
 import type { FetchDataResult } from "../data/fetcher";

+/**
+ * Validates a git branch name against a strict whitelist pattern.
+ * This prevents command injection by ensuring only safe characters are used.
+ *
+ * Valid branch names:
+ * - Start with alphanumeric character (not dash, to prevent option injection)
+ * - Contain only alphanumeric, forward slash, hyphen, underscore, or period
+ * - Do not start or end with a period
+ * - Do not end with a slash
+ * - Do not contain '..' (path traversal)
+ * - Do not contain '//' (consecutive slashes)
+ * - Do not end with '.lock'
+ * - Do not contain '@{'
+ * - Do not contain control characters or special git characters (~^:?*[\])
+ */
+export function validateBranchName(branchName: string): void {
+  // Check for empty or whitespace-only names
+  if (!branchName || branchName.trim().length === 0) {
+    throw new Error("Branch name cannot be empty");
+  }
+
+  // Check for leading dash (prevents option injection like --help, -x)
+  if (branchName.startsWith("-")) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names cannot start with a dash.`,
+    );
+  }
+
+  // Check for control characters and special git characters (~^:?*[\])
+  // eslint-disable-next-line no-control-regex
+  if (/[\x00-\x1F\x7F ~^:?*[\]\\]/.test(branchName)) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names cannot contain control characters, spaces, or special git characters (~^:?*[\\]).`,
+    );
+  }
+
+  // Strict whitelist pattern: alphanumeric start, then alphanumeric/slash/hyphen/underscore/period
+  const validPattern = /^[a-zA-Z0-9][a-zA-Z0-9/_.-]*$/;
+
+  if (!validPattern.test(branchName)) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names must start with an alphanumeric character and contain only alphanumeric characters, forward slashes, hyphens, underscores, or periods.`,
+    );
+  }
+
+  // Check for leading/trailing periods
+  if (branchName.startsWith(".") || branchName.endsWith(".")) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names cannot start or end with a period.`,
+    );
+  }
+
+  // Check for trailing slash
+  if (branchName.endsWith("/")) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names cannot end with a slash.`,
+    );
+  }
+
+  // Check for consecutive slashes
+  if (branchName.includes("//")) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names cannot contain consecutive slashes.`,
+    );
+  }
+
+  // Additional git-specific validations
+  if (branchName.includes("..")) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names cannot contain '..'`,
+    );
+  }
+
+  if (branchName.endsWith(".lock")) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names cannot end with '.lock'`,
+    );
+  }
+
+  if (branchName.includes("@{")) {
+    throw new Error(
+      `Invalid branch name: "${branchName}". Branch names cannot contain '@{'`,
+    );
+  }
+}
+
+/**
+ * Executes a git command safely using execFileSync to avoid shell interpolation.
+ *
+ * Security: execFileSync passes arguments directly to the git binary without
+ * invoking a shell, preventing command injection attacks where malicious input
+ * could be interpreted as shell commands (e.g., branch names containing `;`, `|`, `&&`).
+ *
+ * @param args - Git command arguments (e.g., ["checkout", "branch-name"])
+ */
+function execGit(args: string[]): void {
+  execFileSync("git", args, { stdio: "inherit" });
+}
+
 export type BranchInfo = {
  baseBranch: string;
  claudeBranch?: string;
@@ -53,14 +152,19 @@ export async function setupBranch(
        `PR #${entityNumber}: ${commitCount} commits, using fetch depth ${fetchDepth}`,
      );

+      // Validate branch names before use to prevent command injection
+      validateBranchName(branchName);
+
      // Execute git commands to checkout PR branch (dynamic depth based on PR size)
-      await $`git fetch origin --depth=${fetchDepth} ${branchName}`;
-      await $`git checkout ${branchName} --`;
+      // Using execFileSync instead of shell template literals for security
+      execGit(["fetch", "origin", `--depth=${fetchDepth}`, branchName]);
+      execGit(["checkout", branchName, "--"]);

      console.log(`Successfully checked out PR branch for PR #${entityNumber}`);

      // For open PRs, we need to get the base branch of the PR
      const baseBranch = prData.baseRefName;
+      validateBranchName(baseBranch);

      return {
        baseBranch,
@@ -118,8 +222,9 @@ export async function setupBranch(

      // Ensure we're on the source branch
      console.log(`Fetching and checking out source branch: ${sourceBranch}`);
-      await $`git fetch origin ${sourceBranch} --depth=1`;
-      await $`git checkout ${sourceBranch}`;
+      validateBranchName(sourceBranch);
+      execGit(["fetch", "origin", sourceBranch, "--depth=1"]);
+      execGit(["checkout", sourceBranch, "--"]);

      // Set outputs for GitHub Actions
      core.setOutput("CLAUDE_BRANCH", newBranch);
@@ -138,11 +243,13 @@ export async function setupBranch(

    // Fetch and checkout the source branch first to ensure we branch from the correct base
    console.log(`Fetching and checking out source branch: ${sourceBranch}`);
-    await $`git fetch origin ${sourceBranch} --depth=1`;
-    await $`git checkout ${sourceBranch}`;
+    validateBranchName(sourceBranch);
+    validateBranchName(newBranch);
+    execGit(["fetch", "origin", sourceBranch, "--depth=1"]);
+    execGit(["checkout", sourceBranch, "--"]);

    // Create and checkout the new branch from the source branch
-    await $`git checkout -b ${newBranch}`;
+    execGit(["checkout", "-b", newBranch]);

    console.log(
      `Successfully created and checked out local branch: ${newBranch}`,
--- a/src/github/operations/comment-logic.ts
+++ b/src/github/operations/comment-logic.ts
@@ -1,7 +1,7 @@
 import { GITHUB_SERVER_URL } from "../api/config";

 export type ExecutionDetails = {
-  cost_usd?: number;
+  total_cost_usd?: number;
  duration_ms?: number;
  duration_api_ms?: number;
 };
--- a/src/github/types.ts
+++ b/src/github/types.ts
@@ -58,6 +58,8 @@ export type GitHubPullRequest = {
  headRefName: string;
  headRefOid: string;
  createdAt: string;
+  updatedAt?: string;
+  lastEditedAt?: string;
  additions: number;
  deletions: number;
  state: string;
@@ -83,6 +85,8 @@ export type GitHubIssue = {
  body: string;
  author: GitHubAuthor;
  createdAt: string;
+  updatedAt?: string;
+  lastEditedAt?: string;
  state: string;
  comments: {
    nodes: GitHubComment[];
--- a/test/comment-logic.test.ts
+++ b/test/comment-logic.test.ts
@@ -258,7 +258,7 @@ describe("updateCommentBody", () => {
      const input = {
        ...baseInput,
        executionDetails: {
-          cost_usd: 0.13382595,
+          total_cost_usd: 0.13382595,
          duration_ms: 31033,
          duration_api_ms: 31034,
        },
@@ -301,7 +301,7 @@ describe("updateCommentBody", () => {
      const input = {
        ...baseInput,
        executionDetails: {
-          cost_usd: 0.25,
+          total_cost_usd: 0.25,
        },
        triggerUsername: "testuser",
      };
@@ -322,7 +322,7 @@ describe("updateCommentBody", () => {
        branchName: "claude-branch-123",
        prLink: "\n[Create a PR](https://github.com/owner/repo/pr-url)",
        executionDetails: {
-          cost_usd: 0.01,
+          total_cost_usd: 0.01,
          duration_ms: 65000, // 1 minute 5 seconds
        },
        triggerUsername: "trigger-user",
--- a/test/data-fetcher.test.ts
+++ b/test/data-fetcher.test.ts
@@ -4,6 +4,7 @@ import {
  fetchGitHubData,
  filterCommentsToTriggerTime,
  filterReviewsToTriggerTime,
+  isBodySafeToUse,
 } from "../src/github/data/fetcher";
 import {
  createMockContext,
@@ -371,6 +372,139 @@ describe("filterReviewsToTriggerTime", () => {
  });
 });

+describe("isBodySafeToUse", () => {
+  const triggerTime = "2024-01-15T12:00:00Z";
+
+  const createMockContextData = (
+    createdAt: string,
+    updatedAt?: string,
+    lastEditedAt?: string,
+  ) => ({
+    createdAt,
+    updatedAt,
+    lastEditedAt,
+  });
+
+  describe("body edit time validation", () => {
+    it("should return true when body was never edited", () => {
+      const contextData = createMockContextData("2024-01-15T10:00:00Z");
+      expect(isBodySafeToUse(contextData, triggerTime)).toBe(true);
+    });
+
+    it("should return true when body was edited before trigger time", () => {
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T11:00:00Z",
+        "2024-01-15T11:30:00Z",
+      );
+      expect(isBodySafeToUse(contextData, triggerTime)).toBe(true);
+    });
+
+    it("should return false when body was edited after trigger time (using updatedAt)", () => {
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T13:00:00Z",
+      );
+      expect(isBodySafeToUse(contextData, triggerTime)).toBe(false);
+    });
+
+    it("should return false when body was edited after trigger time (using lastEditedAt)", () => {
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        undefined,
+        "2024-01-15T13:00:00Z",
+      );
+      expect(isBodySafeToUse(contextData, triggerTime)).toBe(false);
+    });
+
+    it("should return false when body was edited exactly at trigger time", () => {
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T12:00:00Z",
+      );
+      expect(isBodySafeToUse(contextData, triggerTime)).toBe(false);
+    });
+
+    it("should prioritize lastEditedAt over updatedAt", () => {
+      // updatedAt is after trigger, but lastEditedAt is before - should be safe
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T13:00:00Z", // updatedAt after trigger
+        "2024-01-15T11:00:00Z", // lastEditedAt before trigger
+      );
+      expect(isBodySafeToUse(contextData, triggerTime)).toBe(true);
+    });
+  });
+
+  describe("edge cases", () => {
+    it("should return true when no trigger time is provided (backward compatibility)", () => {
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T13:00:00Z", // Would normally fail
+        "2024-01-15T14:00:00Z", // Would normally fail
+      );
+      expect(isBodySafeToUse(contextData, undefined)).toBe(true);
+    });
+
+    it("should handle millisecond precision correctly", () => {
+      // Edit 1ms after trigger - should be unsafe
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T12:00:00.001Z",
+      );
+      expect(isBodySafeToUse(contextData, triggerTime)).toBe(false);
+    });
+
+    it("should handle edit 1ms before trigger - should be safe", () => {
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T11:59:59.999Z",
+      );
+      expect(isBodySafeToUse(contextData, triggerTime)).toBe(true);
+    });
+
+    it("should handle various ISO timestamp formats", () => {
+      const contextData1 = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T11:00:00Z",
+      );
+      const contextData2 = createMockContextData(
+        "2024-01-15T10:00:00+00:00",
+        "2024-01-15T11:00:00+00:00",
+      );
+      const contextData3 = createMockContextData(
+        "2024-01-15T10:00:00.000Z",
+        "2024-01-15T11:00:00.000Z",
+      );
+
+      expect(isBodySafeToUse(contextData1, triggerTime)).toBe(true);
+      expect(isBodySafeToUse(contextData2, triggerTime)).toBe(true);
+      expect(isBodySafeToUse(contextData3, triggerTime)).toBe(true);
+    });
+  });
+
+  describe("security scenarios", () => {
+    it("should detect race condition attack - body edited between trigger and processing", () => {
+      // Simulates: Owner triggers @claude at 12:00, attacker edits body at 12:00:30
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z", // Issue created
+        "2024-01-15T12:00:30Z", // Body edited after trigger
+      );
+      expect(isBodySafeToUse(contextData, "2024-01-15T12:00:00Z")).toBe(false);
+    });
+
+    it("should allow body that was stable at trigger time", () => {
+      // Body was last edited well before the trigger
+      const contextData = createMockContextData(
+        "2024-01-15T10:00:00Z",
+        "2024-01-15T10:30:00Z",
+        "2024-01-15T10:30:00Z",
+      );
+      expect(isBodySafeToUse(contextData, "2024-01-15T12:00:00Z")).toBe(true);
+    });
+  });
+});
+
 describe("fetchGitHubData integration with time filtering", () => {
  it("should filter comments based on trigger time when provided", async () => {
    const mockOctokits = {
@@ -696,4 +830,119 @@ describe("fetchGitHubData integration with time filtering", () => {
    // All three comments should be included as they're all before trigger time
    expect(result.comments.length).toBe(3);
  });
+
+  it("should exclude issue body when edited after trigger time (TOCTOU protection)", async () => {
+    const mockOctokits = {
+      graphql: jest.fn().mockResolvedValue({
+        repository: {
+          issue: {
+            number: 555,
+            title: "Test Issue",
+            body: "Malicious body edited after trigger",
+            author: { login: "attacker" },
+            createdAt: "2024-01-15T10:00:00Z",
+            updatedAt: "2024-01-15T12:30:00Z", // Edited after trigger
+            lastEditedAt: "2024-01-15T12:30:00Z", // Edited after trigger
+            comments: { nodes: [] },
+          },
+        },
+        user: { login: "trigger-user" },
+      }),
+      rest: jest.fn() as any,
+    };
+
+    const result = await fetchGitHubData({
+      octokits: mockOctokits as any,
+      repository: "test-owner/test-repo",
+      prNumber: "555",
+      isPR: false,
+      triggerUsername: "trigger-user",
+      triggerTime: "2024-01-15T12:00:00Z",
+    });
+
+    // The body should be excluded from image processing due to TOCTOU protection
+    // We can verify this by checking that issue_body is NOT in the imageUrlMap keys
+    const hasIssueBodyInMap = Array.from(result.imageUrlMap.keys()).some(
+      (key) => key.includes("issue_body"),
+    );
+    expect(hasIssueBodyInMap).toBe(false);
+  });
+
+  it("should include issue body when not edited after trigger time", async () => {
+    const mockOctokits = {
+      graphql: jest.fn().mockResolvedValue({
+        repository: {
+          issue: {
+            number: 666,
+            title: "Test Issue",
+            body: "Safe body not edited after trigger",
+            author: { login: "author" },
+            createdAt: "2024-01-15T10:00:00Z",
+            updatedAt: "2024-01-15T11:00:00Z", // Edited before trigger
+            lastEditedAt: "2024-01-15T11:00:00Z", // Edited before trigger
+            comments: { nodes: [] },
+          },
+        },
+        user: { login: "trigger-user" },
+      }),
+      rest: jest.fn() as any,
+    };
+
+    const result = await fetchGitHubData({
+      octokits: mockOctokits as any,
+      repository: "test-owner/test-repo",
+      prNumber: "666",
+      isPR: false,
+      triggerUsername: "trigger-user",
+      triggerTime: "2024-01-15T12:00:00Z",
+    });
+
+    // The contextData should still contain the body
+    expect(result.contextData.body).toBe("Safe body not edited after trigger");
+  });
+
+  it("should exclude PR body when edited after trigger time (TOCTOU protection)", async () => {
+    const mockOctokits = {
+      graphql: jest.fn().mockResolvedValue({
+        repository: {
+          pullRequest: {
+            number: 777,
+            title: "Test PR",
+            body: "Malicious PR body edited after trigger",
+            author: { login: "attacker" },
+            baseRefName: "main",
+            headRefName: "feature",
+            headRefOid: "abc123",
+            createdAt: "2024-01-15T10:00:00Z",
+            updatedAt: "2024-01-15T12:30:00Z", // Edited after trigger
+            lastEditedAt: "2024-01-15T12:30:00Z", // Edited after trigger
+            additions: 10,
+            deletions: 5,
+            state: "OPEN",
+            commits: { totalCount: 1, nodes: [] },
+            files: { nodes: [] },
+            comments: { nodes: [] },
+            reviews: { nodes: [] },
+          },
+        },
+        user: { login: "trigger-user" },
+      }),
+      rest: jest.fn() as any,
+    };
+
+    const result = await fetchGitHubData({
+      octokits: mockOctokits as any,
+      repository: "test-owner/test-repo",
+      prNumber: "777",
+      isPR: true,
+      triggerUsername: "trigger-user",
+      triggerTime: "2024-01-15T12:00:00Z",
+    });
+
+    // The body should be excluded from image processing due to TOCTOU protection
+    const hasPrBodyInMap = Array.from(result.imageUrlMap.keys()).some((key) =>
+      key.includes("pr_body"),
+    );
+    expect(hasPrBodyInMap).toBe(false);
+  });
 });
--- a/test/fixtures/sample-turns.json
+++ b/test/fixtures/sample-turns.json
@@ -189,7 +189,7 @@
  },
  {
    "type": "result",
-    "cost_usd": 0.0347,
+    "total_cost_usd": 0.0347,
    "duration_ms": 18750,
    "result": "Successfully removed debug print statement from file and added review comment to document the change."
  }
--- a/test/mockContext.ts
+++ b/test/mockContext.ts
@@ -401,6 +401,53 @@ export const mockPullRequestReviewContext: ParsedGitHubContext = {
  inputs: { ...defaultInputs, triggerPhrase: "@claude" },
 };

+export const mockPullRequestReviewWithoutCommentContext: ParsedGitHubContext = {
+  runId: "1234567890",
+  eventName: "pull_request_review",
+  eventAction: "dismissed",
+  repository: defaultRepository,
+  actor: "senior-developer",
+  payload: {
+    action: "submitted",
+    review: {
+      id: 11122233,
+      body: null, // Simulating approval without comment
+      user: {
+        login: "senior-developer",
+        id: 44444,
+        avatar_url: "https://avatars.githubusercontent.com/u/44444",
+        html_url: "https://github.com/senior-developer",
+      },
+      state: "approved",
+      html_url:
+        "https://github.com/test-owner/test-repo/pull/321#pullrequestreview-11122233",
+      submitted_at: "2024-01-15T15:30:00Z",
+    },
+    pull_request: {
+      number: 321,
+      title: "Refactor: Improve error handling in API layer",
+      body: "This PR improves error handling across all API endpoints",
+      user: {
+        login: "backend-developer",
+        id: 33333,
+        avatar_url: "https://avatars.githubusercontent.com/u/33333",
+        html_url: "https://github.com/backend-developer",
+      },
+    },
+    repository: {
+      name: "test-repo",
+      full_name: "test-owner/test-repo",
+      private: false,
+      owner: {
+        login: "test-owner",
+      },
+    },
+  } as PullRequestReviewEvent,
+  entityNumber: 321,
+  isPR: true,
+  inputs: { ...defaultInputs, triggerPhrase: "@claude" },
+};
+
 export const mockPullRequestReviewCommentContext: ParsedGitHubContext = {
  runId: "1234567890",
  eventName: "pull_request_review_comment",
--- a/test/prepare-context.test.ts
+++ b/test/prepare-context.test.ts
@@ -10,6 +10,7 @@ import {
  mockPullRequestCommentContext,
  mockPullRequestReviewContext,
  mockPullRequestReviewCommentContext,
+  mockPullRequestReviewWithoutCommentContext,
 } from "./mockContext";

 const BASE_ENV = {
@@ -126,6 +127,24 @@ describe("parseEnvVarsWithContext", () => {
    });
  });

+  describe("pull_request_review event without comment", () => {
+    test("should parse pull_request_review event correctly", () => {
+      process.env = BASE_ENV;
+      const result = prepareContext(
+        mockPullRequestReviewWithoutCommentContext,
+        "12345",
+      );
+
+      expect(result.eventData.eventName).toBe("pull_request_review");
+      expect(result.eventData.isPR).toBe(true);
+      expect(result.triggerUsername).toBe("senior-developer");
+      if (result.eventData.eventName === "pull_request_review") {
+        expect(result.eventData.prNumber).toBe("321");
+        expect(result.eventData.commentBody).toBe("");
+      }
+    });
+  });
+
  describe("pull_request_review_comment event", () => {
    test("should parse pull_request_review_comment event correctly", () => {
      process.env = BASE_ENV;
--- a/test/validate-branch-name.test.ts
+++ b/test/validate-branch-name.test.ts
@@ -0,0 +1,201 @@
+import { describe, expect, it } from "bun:test";
+import { validateBranchName } from "../src/github/operations/branch";
+
+describe("validateBranchName", () => {
+  describe("valid branch names", () => {
+    it("should accept simple alphanumeric names", () => {
+      expect(() => validateBranchName("main")).not.toThrow();
+      expect(() => validateBranchName("feature123")).not.toThrow();
+      expect(() => validateBranchName("Branch1")).not.toThrow();
+    });
+
+    it("should accept names with hyphens", () => {
+      expect(() => validateBranchName("feature-branch")).not.toThrow();
+      expect(() => validateBranchName("fix-bug-123")).not.toThrow();
+    });
+
+    it("should accept names with underscores", () => {
+      expect(() => validateBranchName("feature_branch")).not.toThrow();
+      expect(() => validateBranchName("fix_bug_123")).not.toThrow();
+    });
+
+    it("should accept names with forward slashes", () => {
+      expect(() => validateBranchName("feature/new-thing")).not.toThrow();
+      expect(() => validateBranchName("user/feature/branch")).not.toThrow();
+    });
+
+    it("should accept names with periods", () => {
+      expect(() => validateBranchName("v1.0.0")).not.toThrow();
+      expect(() => validateBranchName("release.1.2.3")).not.toThrow();
+    });
+
+    it("should accept typical branch name formats", () => {
+      expect(() =>
+        validateBranchName("claude/issue-123-20250101-1234"),
+      ).not.toThrow();
+      expect(() => validateBranchName("refs/heads/main")).not.toThrow();
+      expect(() => validateBranchName("bugfix/JIRA-1234")).not.toThrow();
+    });
+  });
+
+  describe("command injection attempts", () => {
+    it("should reject shell command substitution with $()", () => {
+      expect(() => validateBranchName("$(whoami)")).toThrow();
+      expect(() => validateBranchName("branch-$(rm -rf /)")).toThrow();
+      expect(() => validateBranchName("test$(cat /etc/passwd)")).toThrow();
+    });
+
+    it("should reject shell command substitution with backticks", () => {
+      expect(() => validateBranchName("`whoami`")).toThrow();
+      expect(() => validateBranchName("branch-`rm -rf /`")).toThrow();
+    });
+
+    it("should reject command chaining with semicolons", () => {
+      expect(() => validateBranchName("branch; rm -rf /")).toThrow();
+      expect(() => validateBranchName("test;whoami")).toThrow();
+    });
+
+    it("should reject command chaining with &&", () => {
+      expect(() => validateBranchName("branch && rm -rf /")).toThrow();
+      expect(() => validateBranchName("test&&whoami")).toThrow();
+    });
+
+    it("should reject command chaining with ||", () => {
+      expect(() => validateBranchName("branch || rm -rf /")).toThrow();
+      expect(() => validateBranchName("test||whoami")).toThrow();
+    });
+
+    it("should reject pipe characters", () => {
+      expect(() => validateBranchName("branch | cat")).toThrow();
+      expect(() => validateBranchName("test|grep password")).toThrow();
+    });
+
+    it("should reject redirection operators", () => {
+      expect(() => validateBranchName("branch > /etc/passwd")).toThrow();
+      expect(() => validateBranchName("branch < input")).toThrow();
+      expect(() => validateBranchName("branch >> file")).toThrow();
+    });
+  });
+
+  describe("option injection attempts", () => {
+    it("should reject branch names starting with dash", () => {
+      expect(() => validateBranchName("-x")).toThrow(
+        /cannot start with a dash/,
+      );
+      expect(() => validateBranchName("--help")).toThrow(
+        /cannot start with a dash/,
+      );
+      expect(() => validateBranchName("-")).toThrow(/cannot start with a dash/);
+      expect(() => validateBranchName("--version")).toThrow(
+        /cannot start with a dash/,
+      );
+      expect(() => validateBranchName("-rf")).toThrow(
+        /cannot start with a dash/,
+      );
+    });
+  });
+
+  describe("path traversal attempts", () => {
+    it("should reject double dot sequences", () => {
+      expect(() => validateBranchName("../../../etc")).toThrow();
+      expect(() => validateBranchName("branch/../secret")).toThrow(/'\.\.'$/);
+      expect(() => validateBranchName("a..b")).toThrow(/'\.\.'$/);
+    });
+  });
+
+  describe("git-specific invalid patterns", () => {
+    it("should reject @{ sequence", () => {
+      expect(() => validateBranchName("branch@{1}")).toThrow(/@{/);
+      expect(() => validateBranchName("HEAD@{yesterday}")).toThrow(/@{/);
+    });
+
+    it("should reject .lock suffix", () => {
+      expect(() => validateBranchName("branch.lock")).toThrow(/\.lock/);
+      expect(() => validateBranchName("feature.lock")).toThrow(/\.lock/);
+    });
+
+    it("should reject consecutive slashes", () => {
+      expect(() => validateBranchName("feature//branch")).toThrow(
+        /consecutive slashes/,
+      );
+      expect(() => validateBranchName("a//b//c")).toThrow(
+        /consecutive slashes/,
+      );
+    });
+
+    it("should reject trailing slashes", () => {
+      expect(() => validateBranchName("feature/")).toThrow(
+        /cannot end with a slash/,
+      );
+      expect(() => validateBranchName("branch/")).toThrow(
+        /cannot end with a slash/,
+      );
+    });
+
+    it("should reject leading periods", () => {
+      expect(() => validateBranchName(".hidden")).toThrow();
+    });
+
+    it("should reject trailing periods", () => {
+      expect(() => validateBranchName("branch.")).toThrow(
+        /cannot start or end with a period/,
+      );
+    });
+
+    it("should reject special git refspec characters", () => {
+      expect(() => validateBranchName("branch~1")).toThrow();
+      expect(() => validateBranchName("branch^2")).toThrow();
+      expect(() => validateBranchName("branch:ref")).toThrow();
+      expect(() => validateBranchName("branch?")).toThrow();
+      expect(() => validateBranchName("branch*")).toThrow();
+      expect(() => validateBranchName("branch[0]")).toThrow();
+      expect(() => validateBranchName("branch\\path")).toThrow();
+    });
+  });
+
+  describe("control characters and special characters", () => {
+    it("should reject null bytes", () => {
+      expect(() => validateBranchName("branch\x00name")).toThrow();
+    });
+
+    it("should reject other control characters", () => {
+      expect(() => validateBranchName("branch\x01name")).toThrow();
+      expect(() => validateBranchName("branch\x1Fname")).toThrow();
+      expect(() => validateBranchName("branch\x7Fname")).toThrow();
+    });
+
+    it("should reject spaces", () => {
+      expect(() => validateBranchName("branch name")).toThrow();
+      expect(() => validateBranchName("feature branch")).toThrow();
+    });
+
+    it("should reject newlines and tabs", () => {
+      expect(() => validateBranchName("branch\nname")).toThrow();
+      expect(() => validateBranchName("branch\tname")).toThrow();
+    });
+  });
+
+  describe("empty and whitespace", () => {
+    it("should reject empty strings", () => {
+      expect(() => validateBranchName("")).toThrow(/cannot be empty/);
+    });
+
+    it("should reject whitespace-only strings", () => {
+      expect(() => validateBranchName("   ")).toThrow();
+      expect(() => validateBranchName("\t\n")).toThrow();
+    });
+  });
+
+  describe("edge cases", () => {
+    it("should accept single alphanumeric character", () => {
+      expect(() => validateBranchName("a")).not.toThrow();
+      expect(() => validateBranchName("1")).not.toThrow();
+    });
+
+    it("should reject single special characters", () => {
+      expect(() => validateBranchName(".")).toThrow();
+      expect(() => validateBranchName("/")).toThrow();
+      expect(() => validateBranchName("-")).toThrow();
+    });
+  });
+});
Author	SHA1	Message	Date
Ashwin Bhat	d7b6d50442	fix: merge multiple --mcp-config flags and support --allowed-tools parsing (#748 ) * fix: merge multiple --mcp-config flags instead of overwriting When users provide their own --mcp-config in claude_args, the action's built-in MCP servers (github_comment, github_ci, etc.) were being lost because multiple --mcp-config flags were overwriting each other. This fix: - Adds mcp-config to ACCUMULATING_FLAGS to collect all values - Changes delimiter to null character to avoid conflicts with JSON - Adds mergeMcpConfigs() to combine mcpServers objects from multiple configs - Merges inline JSON configs while preserving file path configs Fixes #745 🤖 Generated with [Claude Code](https://claude.ai/code) Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Ashwin Bhat <ashwin-ant@users.noreply.github.com> * fix: support hyphenated --allowed-tools flag and multiple values The --allowed-tools flag was not being parsed correctly when: 1. Using the hyphenated form (--allowed-tools) instead of camelCase (--allowedTools) 2. Passing multiple space-separated values after a single flag (e.g., --allowed-tools "Tool1" "Tool2" "Tool3") This fix: - Adds hyphenated variants (allowed-tools, disallowed-tools) to ACCUMULATING_FLAGS - Updates parsing to consume all consecutive non-flag values for accumulating flags - Merges values from both camelCase and hyphenated variants Fixes #746 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Ashwin Bhat <ashwin-ant@users.noreply.github.com>	2025-12-16 13:08:25 -08:00
Ashwin Bhat	f375cabfab	chore: update model to claude-opus-4-5 in workflow (#747 ) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude <noreply@anthropic.com>	2025-12-16 12:47:41 -08:00
GitHub Actions	9acae263e7	chore: bump Claude Code to 2.0.70 and Agent SDK to 0.1.70	2025-12-15 23:53:03 +00:00
Gor Grigoryan	67bf0594ce	feat: add session_id output to enable resuming conversations (#739 ) Add a new `session_id` output that exposes the Claude Code session ID, allowing other workflows or Claude Code instances to resume the conversation using `--resume <session_id>`. Changes: - Add parseAndSetSessionId() function to extract session_id from the system.init message in execution output - Add session_id output to both action.yml and base-action/action.yml - Add comprehensive tests for the new functionality 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>	2025-12-14 19:42:54 -08:00
GitHub Actions	b58533dbe0	chore: bump Claude Code version to 2.0.69	2025-12-13 01:00:43 +00:00
GitHub Actions	bda9bf08de	chore: bump Claude Code version to 2.0.68	2025-12-12 23:32:49 +00:00
Ashwin Bhat	79b343c094	feat: Make Agent SDK the default execution path (#738 ) Change USE_AGENT_SDK to default to true instead of false. The Agent SDK path is now used by default; set USE_AGENT_SDK=false to use the CLI path. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude <noreply@anthropic.com>	2025-12-12 13:55:16 -08:00
bogini	609c388361	Fix command injection vulnerability in branch setup (#736 ) * fix: Prevent command injection in branch operations Replace Bun shell template literals with Node.js execFileSync to prevent command injection attacks via malicious branch names. Branch names from PR data (headRefName) are now validated against a strict whitelist pattern before use in git commands. Changes: - Add validateBranchName() function with strict character whitelist - Replace $`git ...` shell templates with execGit() using execFileSync - Validate all branch names before use in git operations * fix: Address review comments for branch validation security - Enhanced execGit JSDoc to explain security benefits of execFileSync - Added comprehensive branch name validation: - Leading dash check (prevents option injection) - Control characters and special git characters (~^:?*[\]) - Leading/trailing period checks - Trailing slash and consecutive slash checks - Added -- separator to git checkout commands - Added 30 unit tests for validateBranchName covering: - Valid branch names - Command injection attempts - Option injection attempts - Path traversal attempts - Git-specific invalid patterns - Control characters and edge cases 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>	2025-12-12 11:30:28 -08:00
GitHub Actions	f0c8eb2980	chore: bump Claude Code version to 2.0.62	2025-12-09 02:12:14 +00:00
ant-soumitr	68a0348c20	fix: Replace direct template expansion of inputs in shell scripts with environment variables (#729 ) Replace direct template expansion of user inputs in shell scripts with environment variables to prevent potential command injection attacks. Changes: - sync-base-action.yml: Use $GITHUB_EVENT_NAME and $GITHUB_ACTOR instead of template expansion - action.yml: Pass path_to_bun_executable and path_to_claude_code_executable through env vars - base-action/action.yml: Same env var changes for path inputs 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude <noreply@anthropic.com>	2025-12-08 12:08:44 -08:00
GitHub Actions	dc06a34646	chore: bump Claude Code version to 2.0.61	2025-12-07 10:47:47 +00:00
Ashwin Bhat	a3bb51dac1	Fix SDK path: add settingSources and default system prompt (#726 ) Two fixes for the Agent SDK path (USE_AGENT_SDK=true): 1. Add settingSources to load filesystem settings - Without this, CLI-installed plugins aren't available to the SDK - Also needed to load CLAUDE.md files from the project 2. Default systemPrompt to claude_code preset - Without an explicit systemPrompt, the SDK would use no system prompt - Now defaults to { type: "preset", preset: "claude_code" } to match CLI behavior Also adds logging of SDK options (excluding env) for debugging. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude <noreply@anthropic.com>	2025-12-06 16:52:26 -08:00
GitHub Actions	6610520549	chore: bump Claude Code version to 2.0.60	2025-12-06 00:10:42 +00:00
GitHub Actions	e2eb96f51d	chore: bump Claude Code version to 2.0.59	2025-12-04 23:09:43 +00:00
Ashwin Bhat	05c95aed79	fix: accumulate multiple --allowedTools flags for Agent SDK (#719 ) * fix: merge allowedTools from claudeArgs when using Agent SDK When USE_AGENT_SDK=true, the allowedTools from claudeArgs (which contains tag mode's required tools like mcp__github_comment__update_claude_comment) were being lost because parseClaudeArgsToExtraArgs converts args to a Record<string, string>, and the SDK was using sdkOptions.allowedTools (from direct options) instead of merging with extraArgs.allowedTools. This fix: - Extracts allowedTools/disallowedTools from extraArgs after parsing - Merges them with any direct options.allowedTools/disallowedTools - Removes them from extraArgs to prevent duplicate CLI flags - Passes the merged list as sdkOptions.allowedTools 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: accumulate multiple --allowedTools flags in claudeArgs When tag mode adds its --allowedTools (with MCP tools) and the user also provides --allowedTools in their claude_args, the parseClaudeArgsToExtraArgs function was only keeping the last value. This caused tag mode's required tools like mcp__github_comment__update_claude_comment to be lost. Now allowedTools and disallowedTools flags accumulate their values when they appear multiple times in claudeArgs, so both tag mode's tools and user's tools are preserved. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>	2025-12-04 10:25:54 -08:00
Ashwin Bhat	bb4a3f68f7	feat: add simplified prompt option via USE_SIMPLE_PROMPT env var (#718 ) Adds a shorter, more concise prompt for tag mode that trusts the model to figure out details. Opt-in via USE_SIMPLE_PROMPT=true. The simplified prompt keeps all context data but reduces instructions from ~250 to ~70 lines. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude <noreply@anthropic.com>	2025-12-04 10:25:47 -08:00
Philippe Laflamme	2acd1f7011	fix: `commentBody` may be `null` (#706 ) * fix: `commentBody` may be `null` This handles the cases where `pull_request_review` events have no comments (`commentBody` field is `null`). In those cases, the `null` value is converted to the empty string. The issue was testing `!commentBody` which was triggerring on empty strings as well. This guard was removed (which is the fix), but for clarity, the `commentBody` field was also made optional to make it clear that the comment may be missing. * fix: bun run format	2025-12-03 17:34:31 -08:00
Ashwin Bhat	469fc9c1a4	feat: add Agent SDK support with USE_AGENT_SDK feature flag (#698 ) * feat: add Agent SDK support with USE_AGENT_SDK feature flag Add a feature-flagged code path that uses the Agent SDK instead of spawning the CLI as a subprocess. When USE_AGENT_SDK=true is set, the new SDK path is used; otherwise, existing CLI behavior is unchanged. Changes: - Add parse-sdk-options.ts for parsing ClaudeOptions into SDK format - Add run-claude-sdk.ts for SDK execution with query() function - Update run-claude.ts with feature flag check at entry point - Update update-comment-link.ts to handle both cost_usd and total_cost_usd - Add @anthropic-ai/claude-agent-sdk dependency 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * refactor: simplify SDK types by using @anthropic-ai/claude-agent-sdk types directly - Remove duplicate SdkRunOptions and McpStdioServerConfig types - Use SDK's Options and McpStdioServerConfig types directly - Return { sdkOptions, showFullOutput, hasJsonSchema } from parseSdkOptions - Remove unnecessary convertMcpServers function - Net reduction of ~70 lines 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * refactor: use extraArgs for claudeArgs pass-through to CLI Simplify option parsing by converting claudeArgs to extraArgs record and letting the SDK/CLI handle --mcp-config, --json-schema, etc. - Remove extractJsonSchema and parseMcpConfigs functions - Add parseClaudeArgsToExtraArgs for simple flag parsing - CLI handles complex args like --mcp-config directly 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * ci * refactor: remove hardcoded permission bypass flags The SDK path should match CLI path behavior - permissions are handled by the CLI itself, not hardcoded in the action. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * chore: add logging for SDK vs CLI path selection --------- Co-authored-by: Claude <noreply@anthropic.com>	2025-12-03 17:22:04 -08:00
GitHub Actions	90da6b6e15	chore: bump Claude Code version to 2.0.58	2025-12-03 20:09:55 +00:00
GitHub Actions	752ba96ea1	chore: bump Claude Code version to 2.0.57	2025-12-03 05:24:27 +00:00
GitHub Actions	66bf95c07f	chore: bump Claude Code version to 2.0.56	2025-12-02 01:35:17 +00:00
Ashwin Bhat	6337623ebb	fix: prevent TOCTOU race condition on issue/PR body edits (#710 ) Add trigger-time validation for issue/PR body content to prevent attackers from exploiting a race condition where they edit the body between when an authorized user triggers @claude and when Claude processes the request. The existing filterCommentsToTriggerTime() already protected comments - this extends the same pattern to the main issue/PR body via isBodySafeToUse(). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude <noreply@anthropic.com>	2025-12-01 07:59:39 -08:00
GitHub Actions	6d79044f1d	chore: bump Claude Code version to 2.0.55	2025-11-27 00:01:22 +00:00