diff --git a/.github/workflows/presubmit.yml b/.github/workflows/presubmit.yml
index e22692f352..2c140f8e24 100644
--- a/.github/workflows/presubmit.yml
+++ b/.github/workflows/presubmit.yml
@@ -178,6 +178,133 @@ jobs:
                 -DVULKAN_LIB_DIR='${{ github.workspace }}'/Vulkan-Loader/build/loader/ \
                 ${CMAKE_ADDITIONAL_CONFIG_ARGS}
           cmake --build . --parallel
+
+  tests:
+    name: Run Tests (pocl)
+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v6
+      - name: Parse triggers
+        id: check_triggers
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          MESSAGES=$(gh api -X GET /repos/${{ github.repository }}/pulls/${{ github.event.pull_request.number }}/commits --jq '.[].commit.message')
+          PR_BODY=$(gh api -X GET /repos/${{ github.repository }}/pulls/${{ github.event.pull_request.number }} --jq '.body')
+          COMMANDS=$(echo -e "$MESSAGES\n$PR_BODY" | grep -oP '^\[run-tests?:\s*\K[^\]]+' | sort -u || true)
+          if [ -z "$COMMANDS" ]; then
+            echo "No '[run-test: ...]' triggers found. Skipping job."
+            echo "run_tests=false" >> $GITHUB_OUTPUT
+          else
+            echo "Found the following test triggers:"
+            echo "$COMMANDS"
+            echo "run_tests=true" >> $GITHUB_OUTPUT
+            echo "$COMMANDS" > test_commands.txt
+          fi
+      - name: Environment variables
+        if: steps.check_triggers.outputs.run_tests == 'true'
+        run: |
+          echo "pocl=$(realpath ${{ github.workspace }}/../pocl)" >> $GITHUB_ENV
+          echo "loader=$(realpath ${{ github.workspace }}/../loader)" >> $GITHUB_ENV
+          echo "cl-headers=$(realpath ${{ github.workspace }}/../cl-headers)" >> $GITHUB_ENV
+          echo "spv-headers=$(realpath ${{ github.workspace }}/../spv-headers)" >> $GITHUB_ENV
+      - name: Clone Headers
+        if: steps.check_triggers.outputs.run_tests == 'true'
+        run: |
+          git clone --depth 1 https://github.com/KhronosGroup/OpenCL-Headers.git ${{ env.cl-headers }}
+          git clone --depth 1 https://github.com/KhronosGroup/SPIRV-Headers.git ${{ env.spv-headers }}
+      - name: Setup Ninja
+        uses: seanmiddleditch/gha-setup-ninja@master
+      - name: Install Vulkan SDK
+        uses: humbletim/install-vulkan-sdk@main
+        with:
+          version: 1.4.309.0
+          cache: true
+      - name: Install pocl dependencies
+        if: steps.check_triggers.outputs.run_tests == 'true'
+        run: |
+          sudo apt-get update
+          export LLVM_VERSION=20
+          sudo apt-get install -y \
+            python3-dev libpython3-dev build-essential ocl-icd-libopencl1 \
+            cmake git pkg-config libclang-${LLVM_VERSION}-dev clang-${LLVM_VERSION} \
+            llvm-${LLVM_VERSION} make ninja-build ocl-icd-libopencl1 ocl-icd-dev \
+            ocl-icd-opencl-dev libhwloc-dev zlib1g zlib1g-dev clinfo dialog apt-utils \
+            libxml2-dev libclang-cpp${LLVM_VERSION}-dev libclang-cpp${LLVM_VERSION} \
+            llvm-${LLVM_VERSION}-dev
+      - name: Cache pocl
+        id: cache-pocl
+        if: steps.check_triggers.outputs.run_tests == 'true'
+        uses: actions/cache@v4
+        with:
+          path: ${{ env.pocl }}
+          key: ubuntu-24.04-pocl-v7.1
+      - name: Build pocl
+        if: steps.check_triggers.outputs.run_tests == 'true' && steps.cache-pocl.outputs.cache-hit != 'true'
+        run: |
+          git clone --depth 1 --branch v7.1 https://github.com/pocl/pocl.git
+          cmake -B pocl/build -S pocl -G Ninja \
+            -DCMAKE_INSTALL_PREFIX=${{ env.pocl }} \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DENABLE_ICD=ON
+          cmake --build pocl/build
+          cmake --install pocl/build
+      - name: Build loader
+        if: steps.check_triggers.outputs.run_tests == 'true'
+        shell: bash
+        run: |
+          git clone --depth 1 https://github.com/KhronosGroup/OpenCL-ICD-Loader.git
+          cmake -B OpenCL-ICD-Loader/build -S OpenCL-ICD-Loader -G Ninja \
+            -DCMAKE_INSTALL_PREFIX=${{ env.loader }} \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DOPENCL_ICD_LOADER_HEADERS_DIR='${{ env.cl-headers }}'
+          cmake --build OpenCL-ICD-Loader/build
+          cmake --install OpenCL-ICD-Loader/build
+      - name: Build OpenCL CTS
+        if: steps.check_triggers.outputs.run_tests == 'true'
+        shell: bash
+        run: |
+          cmake -B build -S . -G Ninja \
+                -DCMAKE_BUILD_TYPE=Release \
+                -DCL_INCLUDE_DIR='${{ env.cl-headers }}' \
+                -DCL_LIB_DIR=${{ env.loader }}/lib \
+                -DSPIRV_INCLUDE_DIR='${{ env.spv-headers }}' \
+                -DCMAKE_RUNTIME_OUTPUT_DIRECTORY='${{ github.workspace }}'/bin \
+                -DOPENCL_LIBRARIES="-lOpenCL -lpthread" \
+                -DUSE_CL_EXPERIMENTAL=ON \
+                -DGL_IS_SUPPORTED=OFF \
+                -DVULKAN_IS_SUPPORTED=OFF
+          cmake --build build
+      - name: Run tests
+        if: steps.check_triggers.outputs.run_tests == 'true'
+        shell: bash
+        run: |
+          export OCL_ICD_ENABLE_TRACE=1
+          export OCL_ICD_VENDORS=${{ env.pocl }}/etc/OpenCL/vendors
+          export LD_LIBRARY_PATH=${{ env.loader }}/lib:$LD_LIBRARY_PATH
+          mkdir -p test_results
+          while IFS= read -r cmd; do
+            if [ -n "$cmd" ]; then
+              SAFE_NAME=$(echo "$cmd" | sed -e 's/[^A-Za-z0-9._-]/_/g')
+              RESULT_FILE="test_results/${SAFE_NAME}.json"
+              export CL_CONFORMANCE_RESULTS_FILENAME="$RESULT_FILE"
+              echo "=================================================="
+              echo "Executing: ./$cmd"
+              echo "Writing JSON result to: $RESULT_FILE"
+              echo "=================================================="
+              ./bin/$cmd || true
+              cat ${RESULT_FILE}
+            fi
+          done < test_commands.txt
+      - name: Check results
+        if: steps.check_triggers.outputs.run_tests == 'true'
+        shell: bash
+        run: |
+          ls test_results
+          python3 ./ci/compare_results.py \
+            --golden ./ci/pocl/golden.json \
+            --results-dir test_results/
+
   formatcheck:
     name: Check code format
     runs-on: ubuntu-22.04
diff --git a/README.md b/README.md
index 77e4d9a9bd..0b5eabad49 100644
--- a/README.md
+++ b/README.md
@@ -130,3 +130,24 @@ PRs to the repository are required to be `clang-format` clean to pass CI.
 Developers can either use the `git-clang-format` tool locally to verify this
 before contributing, or update their PR based on the diff provided by a failing
 CI job.
+
+## Running Targeted CI Tests on Pull Requests
+
+To help verify fixes or check for regressions without running the entire
+conformance test suite, our continuous integration pipeline allows contributor
+to trigger specific tests on Pull Requests against the `pocl` implementation.
+
+### How to Trigger Tests
+
+Testing is triggered by adding a special tag to either your
+**Pull Request description** or in any of your **commit messages**.
+
+The CI parses the text for the following syntax:
+`[run-test: <command>]`
+
+Multiples tags for a single Pull Request is supported.
+
+### Examples
+
+```text
+[run-test: test_bruteforce exp -w -1]
diff --git a/ci/compare_results.py b/ci/compare_results.py
new file mode 100644
index 0000000000..288c4c380d
--- /dev/null
+++ b/ci/compare_results.py
@@ -0,0 +1,132 @@
+#!/usr/bin/env python3
+import argparse
+import json
+import os
+import sys
+from pathlib import Path
+
+# Terminal Color Codes
+RED = '\033[91m'
+GREEN = '\033[92m'
+YELLOW = '\033[93m'
+RESET = '\033[0m'
+
+def parse_arguments():
+    """Parse command line arguments."""
+    parser = argparse.ArgumentParser(description="Compare PR test results against a golden reference.")
+    parser.add_argument("--golden", required=True, help="Path to golden reference JSON")
+    parser.add_argument("--results-dir", required=True, help="Directory containing PR test JSON results")
+    return parser.parse_args()
+
+def load_golden_reference(filepath):
+    """Load and return the golden reference JSON data."""
+    try:
+        with open(filepath, 'r') as f:
+            return json.load(f)
+    except Exception as e:
+        print(f"Failed to load golden reference '{filepath}': {e}")
+        sys.exit(1)
+
+def categorize_difference(cmd, sub_test, expected, actual):
+    """Generate a formatted message based on the type of difference."""
+    if actual == "pass":
+        return f"{GREEN} FIX: [{cmd}] '{sub_test}' expected '{expected}', but got '{actual}'.{RESET}"
+    elif actual == "fail":
+        return f"{RED} REGRESSION: [{cmd}] '{sub_test}' expected '{expected}', but got '{actual}'.{RESET}"
+    else:
+        return f"{YELLOW} DIFFERENCE: [{cmd}] '{sub_test}' expected '{expected}', but got '{actual}'.{RESET}"
+
+def compare_test_subset(cmd, actual_results, expected_results, differences, missing_refs):
+    """Compare the run subset against expectations. Mutates lists and returns error status."""
+    has_error = False
+
+    for sub_test, actual_status in actual_results.items():
+        # Rule 1: Everything run must exist in the reference
+        if sub_test not in expected_results:
+            missing_refs.append(f"[{cmd}] Sub-test '{sub_test}' not found in golden reference.")
+            has_error = True
+        else:
+            expected_status = expected_results[sub_test]
+            # Rule 2: Note ANY difference
+            if actual_status != expected_status:
+                differences.append(categorize_difference(cmd, sub_test, expected_status, actual_status))
+                has_error = True
+    return has_error
+
+def process_all_results(results_dir, golden_data):
+    """Iterate through all result JSONs and compare them. Returns aggregate data."""
+    has_error = False
+    differences = []
+    missing_refs = []
+
+    if len(list(Path(results_dir).glob("*.json"))) == 0:
+        print(f"Error: Results directory '{results_dir}' is empty.")
+        sys.exit(1)
+    for result_file in Path(results_dir).glob("*.json"):
+        with open(result_file, 'r') as f:
+            try:
+                data = json.load(f)
+            except json.JSONDecodeError:
+                print(f"Error parsing {result_file}. Skipping.")
+                has_error = True
+                continue
+
+        cmd = os.path.basename(data.get("cmd"))
+        results = data.get("results", {})
+
+        if not cmd:
+            print(f"File {result_file} is missing the 'cmd' key. Skipping.")
+            has_error = True
+            continue
+
+        if cmd not in golden_data:
+            missing_refs.append(f"cmd '{cmd}' not found in golden reference.")
+            has_error = True
+            continue
+
+        # Check the specific results against the reference
+        subset_error = compare_test_subset(cmd, results, golden_data[cmd], differences, missing_refs)
+        if subset_error:
+            has_error = True
+
+    return has_error, differences, missing_refs
+
+def print_report_and_exit(has_error, differences, missing_refs):
+    """Print the final comparison report and exit with the appropriate status code."""
+    if missing_refs:
+        print("\n--- Missing References ---")
+        for msg in missing_refs:
+            print(msg)
+        print("\nPlease update the golden reference to include these missing cmd/tests.")
+
+    if differences:
+        print("\n--- Test Differences ---")
+        for diff in differences:
+            print(diff)
+
+    if has_error:
+        print(f"\n{RED} Errors found during comparison. Failing the check.{RESET}")
+        sys.exit(1)
+    else:
+        print(f"\n{GREEN} All run tests match the golden reference perfectly!{RESET}")
+        sys.exit(0)
+
+def main():
+    """Main execution flow."""
+    args = parse_arguments()
+
+    results_dir = Path(args.results_dir)
+    if not results_dir.is_dir():
+        print(f"Error: Results directory '{args.results_dir}' does not exist.")
+        sys.exit(1)
+
+    golden_data = load_golden_reference(args.golden)
+
+    # Process the files and gather all differences without stopping early
+    has_error, differences, missing_refs = process_all_results(results_dir, golden_data)
+
+    # Output the findings and exit
+    print_report_and_exit(has_error, differences, missing_refs)
+
+if __name__ == "__main__":
+    main()
diff --git a/ci/pocl/golden.json b/ci/pocl/golden.json
new file mode 100644
index 0000000000..0187f311d8
--- /dev/null
+++ b/ci/pocl/golden.json
@@ -0,0 +1,9 @@
+{
+    "test_computeinfo": {
+        "computeinfo": "pass",
+        "device_uuid": "skip",
+        "extended_versioning": "pass",
+        "conformance_version": "pass",
+        "pci_bus_info": "skip"
+    }
+}