diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 0f00aff..34bb55f 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -18,7 +18,9 @@ jobs:
       - uses: actions/setup-python@8d9ed9ac5c53483de85588cdf95a591a75ab9f55 # v5.5.0
         with:
           python-version: '3.13'
-      - uses: pre-commit/action@2c7b3805fd2a0fd8c1884dcaebf91fc102a13ecd # v3.0.1
+      - run: pip install pre-commit                                                                                                                                                                                                
+      - run: pre-commit run --all-files --show-diff-on-failure --color=always
+  
 
   test:
     runs-on: ubuntu-latest
@@ -47,3 +49,18 @@ jobs:
         run: pip install zizmor
       - name: Run zizmor
         run: zizmor --min-severity medium .github/
+
+  security:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: write
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          persist-credentials: false
+      - uses: ./
+        with:
+          bandit_scan_dirs: src
+          package_manager: uv
+          post_pr_comment: ${{ github.event_name == 'pull_request' }}
diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 0000000..f9f5641
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,74 @@
+# CLAUDE.md — python-security-auditing
+
+## What This Project Is
+
+A **GitHub Action** that runs **bandit** (static code analysis) and **pip-audit** (dependency vulnerability scanning) on Python repos, then consolidates results into a single PR comment, workflow step summary, and downloadable artifact.
+
+It is a composite action (`action.yml`) backed by a small Python package (`src/python_security_auditing/`).
+
+## Architecture
+
+```
+action.yml                  ← GitHub Action entry point (composite steps)
+src/python_security_auditing/
+  __main__.py               ← Orchestrator: settings → runners → report → comment → exit
+  settings.py               ← Pydantic-based config from env vars (GitHub Action inputs)
+  runners.py                ← Tool invocation: SARIF parsing, pip-audit, package manager adapters
+  report.py                 ← Markdown report builder and threshold checker
+  pr_comment.py             ← Upsert PR comment via `gh` CLI
+```
+
+**Flow:** `Settings` loads env vars → `runners` invokes tools and parses output → `report` builds markdown and checks thresholds → `pr_comment` posts/updates the PR comment → `__main__` exits 0 or 1.
+
+**Key boundaries:**
+- `settings.py` — input/config boundary (reads env vars, validates via Pydantic)
+- `runners.py` — external tool boundary (subprocess calls to bandit SARIF, pip-audit, package managers)
+- `report.py` — pure logic (markdown generation, threshold checking — no I/O except step summary)
+- `pr_comment.py` — GitHub API boundary (subprocess calls to `gh` CLI)
+
+## Build & Dev
+
+- **Build system:** Hatch (`hatchling`)
+- **Python:** ≥ 3.13
+- **Dependencies:** `pydantic-settings`, `pip-audit`
+- **Dev deps:** `pytest`, `pytest-mock`, `mypy` (strict), `ruff`
+
+### Common Commands
+
+```bash
+# Install in dev mode
+uv pip install -e ".[dev]"
+
+# Run tests
+uv run pytest
+
+# Type checking (strict mode)
+uv run mypy src/
+
+# Lint and format
+uv run ruff check src/ tests/
+uv run ruff format src/ tests/
+```
+
+## Testing Conventions
+
+- Tests live in `tests/` and mirror module names: `test_settings.py`, `test_runners.py`, `test_report.py`.
+- Test fixtures (JSON/SARIF samples) are in `tests/fixtures/`.
+- External tool calls (`subprocess.run`) are mocked in tests — never invoke real `bandit`, `pip-audit`, or `gh` in unit tests.
+- Settings are configured via `monkeypatch.setenv()` since `Settings` reads from env vars.
+- Run the full suite after any change: `pytest`.
+
+## Code Style
+
+- **Formatter/linter:** Ruff (line-length 100, rules: E, F, I, UP)
+- **Type checking:** mypy strict mode
+- **Imports:** `from __future__ import annotations` in every module
+- **Type annotations:** use `dict[str, Any]`, `list[...]`, `int | None` (modern syntax, no `Optional`/`Dict`/`List`)
+- Match existing patterns — don't refactor surrounding code when making a change.
+
+## Key Design Decisions
+
+- **SARIF input for bandit:** Bandit runs in a separate composite step (`lhoupert/bandit-action`). This package only reads the SARIF output file — it does not invoke bandit directly.
+- **PR comment is idempotent:** Uses a hidden HTML marker (`<!-- security-scan-results -->`) to find and update the same comment on subsequent pushes.
+- **Threshold logic:** `check_thresholds()` in `report.py` returns a boolean; the orchestrator translates that to `sys.exit(1)`.
+- **Package manager adapters:** `generate_requirements()` normalizes all package managers to a `requirements.txt` file before passing to `pip-audit`.
diff --git a/action.yml b/action.yml
index 7138bd7..45832b7 100644
--- a/action.yml
+++ b/action.yml
@@ -41,15 +41,11 @@ runs:
         targets: ${{ inputs.bandit_scan_dirs }}
         level: ${{ inputs.bandit_severity_threshold }}
 
-    - name: Set up Python
-      uses: actions/setup-python@8d9ed9ac5c53483de85588cdf95a591a75ab9f55 # v5.5.0
+    - name: Set up uv
+      uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7.6.0
       with:
         python-version: '3.13'
 
-    - name: Install python-security-auditing
-      shell: bash
-      run: pip install "${{ github.action_path }}"
-
     - name: Run security audit
       id: audit
       shell: bash
@@ -65,11 +61,11 @@ runs:
         POST_PR_COMMENT: ${{ inputs.post_pr_comment }}
         GITHUB_TOKEN: ${{ inputs.github_token }}
         PR_NUMBER: ${{ github.event.pull_request.number }}
-      run: python -m python_security_auditing
+      run: uv run --with "${{ github.action_path }}" python -m python_security_auditing
 
     - name: Upload audit reports
       if: always()
-      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+      uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7
       with:
         name: security-audit-reports
         path: |
diff --git a/src/python_security_auditing/runners.py b/src/python_security_auditing/runners.py
index 44110ee..d39581d 100644
--- a/src/python_security_auditing/runners.py
+++ b/src/python_security_auditing/runners.py
@@ -119,7 +119,10 @@ def run_pip_audit(requirements_path: Path) -> list[dict[str, Any]]:
 
     raw = result.stdout.strip()
     if raw:
-        parsed: list[dict[str, Any]] = json.loads(raw)
+        parsed: Any = json.loads(raw)
         output_file.write_text(raw)
-        return parsed
+        # pip-audit 2.7+ wraps output in {"dependencies": [...], "fixes": [...]}
+        if isinstance(parsed, dict):
+            return list(parsed.get("dependencies", []))
+        return list(parsed)
     return []
diff --git a/tests/test_runners.py b/tests/test_runners.py
index 4accb3e..a1effbb 100644
--- a/tests/test_runners.py
+++ b/tests/test_runners.py
@@ -148,12 +148,15 @@ def test_read_bandit_sarif_falls_back_to_level_mapping(tmp_path: Path) -> None:
 
 def test_run_pip_audit_parses_json(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
     monkeypatch.chdir(tmp_path)
-    fixture_text = (FIXTURES / "pip_audit_fixable.json").read_text()
+    # pip-audit 2.7+ wraps output in {"dependencies": [...], "fixes": [...]}
+    deps = json.loads((FIXTURES / "pip_audit_fixable.json").read_text())
+    fixture_text = json.dumps({"dependencies": deps, "fixes": []})
 
     with patch("python_security_auditing.runners.subprocess.run") as mock_run:
         mock_run.return_value = MagicMock(returncode=1, stderr="", stdout=fixture_text)
         report = run_pip_audit(Path("requirements.txt"))
 
+    assert isinstance(report, list)
     assert len(report) == 2
     assert report[0]["name"] == "requests"
     assert (tmp_path / "pip-audit-report.json").exists()