argus-ai-scanner 1.1.0__tar.gz

argus_ai_scanner-1.1.0/.gitignore

@@ -0,0 +1,60 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+*.egg-info/
+.eggs/
+build/
+dist/
+
+# uv / venvs
+.venv/
+venv/
+.python-version
+uv.lock
+
+# Env / secrets
+.env
+.env.*
+!.env.example
+*.pem
+*.key
+
+# Logs / runtime
+logs/
+*.log
+journals/
+results/
+bench_results/*
+!bench_results/v1_1_launch/
+bench_results/v1_1_launch/*
+!bench_results/v1_1_launch/launch_report.md
+_eval_*.json
+_phase_*_evaluation.json
+_full_regression_*.json
+_baseline_run_*.json
+
+# IDE
+.vscode/
+.idea/
+*.swp
+.DS_Store
+
+# Frontend
+node_modules/
+frontend/dist/
+frontend/.vite/
+
+# Test
+.pytest_cache/
+.coverage
+htmlcov/
+.mypy_cache/
+.ruff_cache/
+
+# Local-only
+TODO.local.md
+NOTES.local.md
+.claude/settings.local.json

argus_ai_scanner-1.1.0/CONTRIBUTING.md

@@ -0,0 +1,94 @@
+# Contributing to Argus
+
+Thanks for considering a contribution. Argus is an AI-native code security scanner; the things we care most about are detection quality, cost discipline, and DAST verification fidelity. PRs are welcome.
+
+## Quick start
+
+```bash
+git clone git@github.com:dshochat/Argus_Scanner.git
+cd Argus_Scanner
+uv sync --extra dev
+cp .env.example .env       # add ANTHROPIC_API_KEY + GEMINI_API_KEY
+uv run pytest tests/unit -v
+uv run argus scan path/to/file.py
+```
+
+You'll need:
+
+- Python 3.12+
+- An Anthropic API key (Sonnet 4.6 / Opus 4.6 cascade tiers)
+- A Google AI Studio key (Gemini Flash-Lite triage)
+- Optionally, a Fly.io token for the DAST sandbox tier — not required for most contributions
+
+## Where to start
+
+- **Good first issues** are tagged in [GitHub Issues](https://github.com/dshochat/Argus_Scanner/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22). They're scoped, clearly described, and reviewable in a sitting.
+- **The architecture overview is [`docs/architecture.md`](docs/architecture.md).** Read this before touching `scanner/engine.py` or `dast/orchestrator.py` — both are integration glue with non-obvious invariants.
+
+## Style
+
+- **Python 3.12+, mypy `--strict`, ruff for lint + format.** No flake8 or black; please don't introduce them.
+- **Pydantic v2** for every cross-boundary structure.
+- **`structlog`** for logging; never `print()` outside the CLI.
+- **Type everything.** `Any` is allowed at runner-injection seams (engine.py uses `Any` for `*_runner` parameters by design — duck typing is the testability story); elsewhere prefer concrete types.
+
+```bash
+uv run ruff check . && uv run ruff format .
+uv run mypy --strict .
+```
+
+CI runs both on every PR.
+
+## Tests
+
+Two tiers, kept strictly separate:
+
+- **`tests/unit/`** — no live API. Stubs all model calls. Must pass on every PR. Several hundred tests; fast (<3s for the full suite).
+- **`tests/integration/`** — live API calls. Marked with `@pytest.mark.integration`. Skips if API keys aren't set. Costs API credits on every run. **CI does not run these** — run them locally before submitting cascade or runner changes.
+
+```bash
+uv run pytest tests/unit -v                  # always
+uv run pytest tests/integration -v -s        # before runner / engine changes
+```
+
+Don't add live API calls to `tests/unit/`. The rule has no exceptions: it's the difference between "dev iteration is fast" and "every contributor needs to fund their own scan budget."
+
+## What we welcome
+
+- **New deterministic detectors** in `preprocessing/` — fast, free, well-tested. Mirror existing detectors (`crypto_sensitivity.py`, `imperative_install.py`) for shape.
+- **Prompt tuning** that improves verdict-distance on the regression suite (`samples/regression_v1/`). Show the before/after numbers in the PR.
+- **DAST coverage extensions** — new sandbox image variants, new payload templates for Discovery mode, new oracle types.
+- **Cost guardrails** — anything that makes per-scan cost more predictable or transparent.
+- **Bug reports + reproductions** — issues with a reduced test case land fastest.
+
+## What's out of scope right now
+
+- Adding new model providers beyond Anthropic + Google — defer until v2 benchmark mode resurrects.
+- Architectural pivots without prior discussion in an issue.
+- Frontend work — `frontend/` is dormant for the deferred hosted tier.
+
+## Architecture invariants (non-negotiable)
+
+If your change touches one of these, expect a longer review:
+
+1. **Preprocessing is deterministic and free.** Never call models in `preprocessing/`. If you're tempted, the change probably belongs in `analysis/`.
+2. **The cascade short-circuits cheap files cheap.** Clean files cost $0.0001 (triage only); don't add expensive defaults.
+3. **All runners are injectable.** `scan_file(triage_runner=, sonnet_runner=, opus_runner=, dast_runner=)`. Never hard-code provider calls in the engine.
+4. **DAST never silently lowers an L1 verdict.** A `malicious` → `suspicious` downgrade only fires when *every* L1 finding is sandbox-grounded as `BLOCKED` or `UNREACHED`. Without that, L1's verdict stands.
+5. **Methodology before lift claims.** Don't publish a verdict-exact number from a single regression run. N=2 minimum for cross-config comparisons.
+
+## Pull request process
+
+1. **Open an issue first** if the change is non-trivial. Saves rework on both sides.
+2. **Branch from `main`.** Keep PRs focused — one task ID per PR is the norm.
+3. **Tests pass.** `uv run pytest tests/unit -v && uv run ruff check . && uv run mypy --strict .`
+4. **Live integration smoke** for runner / engine changes — paste the resulting verdict + cost in the PR description.
+5. **Squash-merge by default.** Multi-commit PRs are fine if the commits tell a coherent story.
+
+## Security disclosures
+
+Please do not file public issues for security vulnerabilities. See [SECURITY.md](SECURITY.md) for the private disclosure path.
+
+## License
+
+By contributing, you agree that your contributions will be licensed under the [Apache License 2.0](LICENSE).

argus_ai_scanner-1.1.0/LICENSE

@@ -0,0 +1,190 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for describing the origin of the Work and
+      reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may accept and charge a
+      fee for, or grant additional rights such as warranty, support,
+      indemnity, or other liability obligations and/or rights consistent
+      with this License. However, in accepting such obligations, You may
+      act only on Your own behalf and on Your sole responsibility, not
+      on behalf of any other Contributor, and only if You agree to
+      indemnify, defend, and hold each Contributor harmless for any
+      liability incurred by, or claims asserted against, such Contributor
+      by reason of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   Copyright 2026 Dudy Shochat
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

argus_ai_scanner-1.1.0/PKG-INFO

@@ -0,0 +1,315 @@
+Metadata-Version: 2.4
+Name: argus-ai-scanner
+Version: 1.1.0
+Summary: AI-native code security scanner with cascade analysis and Firecracker-microVM DAST runtime validation
+Project-URL: Homepage, https://github.com/dshochat/Argus_Scanner
+Project-URL: Documentation, https://github.com/dshochat/Argus_Scanner#readme
+Project-URL: Repository, https://github.com/dshochat/Argus_Scanner
+Project-URL: Issues, https://github.com/dshochat/Argus_Scanner/issues
+Project-URL: Changelog, https://github.com/dshochat/Argus_Scanner/releases
+Author-email: David Shochat <davidsho1131@gmail.com>
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: ai-security,anthropic,claude,code-scanner,dast,prompt-injection,sarif,security,static-analysis,supply-chain-security,vulnerability-scanner
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Information Technology
+Classifier: Intended Audience :: System Administrators
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Security
+Classifier: Topic :: Software Development :: Quality Assurance
+Classifier: Topic :: Software Development :: Testing
+Classifier: Typing :: Typed
+Requires-Python: >=3.12
+Requires-Dist: anthropic>=0.40.0
+Requires-Dist: google-genai>=0.3.0
+Requires-Dist: httpx>=0.27.0
+Requires-Dist: pathspec>=0.12.0
+Requires-Dist: pydantic>=2.8.0
+Requires-Dist: python-dotenv>=1.0.0
+Requires-Dist: pyyaml>=6.0.0
+Requires-Dist: requests>=2.32.0
+Requires-Dist: structlog>=24.4.0
+Requires-Dist: tiktoken>=0.7.0
+Provides-Extra: dev
+Requires-Dist: mypy>=1.11.0; extra == 'dev'
+Requires-Dist: pytest-asyncio>=0.24.0; extra == 'dev'
+Requires-Dist: pytest-cov>=5.0.0; extra == 'dev'
+Requires-Dist: pytest>=8.3.0; extra == 'dev'
+Requires-Dist: respx>=0.21.0; extra == 'dev'
+Requires-Dist: ruff>=0.6.0; extra == 'dev'
+Provides-Extra: docs
+Requires-Dist: mkdocs-material>=9.5.0; extra == 'docs'
+Requires-Dist: mkdocs>=1.6.0; extra == 'docs'
+Description-Content-Type: text/markdown
+
+# Argus
+
+**An AI-native code security scanner (Semantic Deep Analysis) <mark>that proves exploitability at runtime</mark>.**
+
+Argus combines a cost-graduated LLM cascade (Gemini Flash-Lite → Sonnet 4.6 → Opus 4.6) with a sandbox tier that *executes* suspect code in a Firecracker microVM and observes what it actually does. Static-analysis findings get promoted to **CONFIRMED** only when the sandbox captures concrete runtime evidence — a network call, a file write, a process spawn. Findings that cannot be triggered are marked **UNREACHED**; findings the file's own defenses block are **BLOCKED**. No more "the LLM said it might be malicious."
+
+Open source, Apache 2.0, BYOK. You pay your providers directly — Anthropic + Google for the cascade, Fly.io for the optional DAST sandbox. Argus collects nothing.
+
+---
+
+## What makes it different
+
+Most scanners stop at "this code matches a vulnerability pattern." Argus runs the code, watches what it does, and reports per-finding outcomes:
+
+| Status | Meaning |
+|---|---|
+| `CONFIRMED` | The sandbox observed the exploit firing at runtime. PoC + event trace are surfaced with the finding. |
+| `BLOCKED` | The attack was tested; the file's own code defended against it (sanitization, escaping, allowlist, etc.). |
+| `UNREACHED` | The attack was tested; the code path is genuinely unreachable. |
+| `NOT_TESTED` | Sandbox couldn't execute the test (with a sub-reason: `infra_stub` / `inconclusive` / `not_planned`). |
+
+### vs. other approaches
+
+| Approach | Output | False-positive burden | Evidence type |
+|---|---|---|---|
+| Pattern-match scanner (regex / AST) | Syntactic match | High | None |
+| Single frontier LLM (single-call) | Probabilistic opinion (semantic) | Medium-high | LLM reasoning |
+| **Argus** | **Runtime-verified verdict** | **Low** | **Sandbox traces** |
+
+A `CONFIRMED` finding looks like this in `argus scan` output:
+
+```json
+{
+  "cwe": "CWE-200",
+  "type": "data_exfiltration",
+  "severity": "critical",
+  "status": "CONFIRMED",
+  "confidence": 1.0,
+  "runtime_evidence": "Mock HTTP server at 127.0.0.1:8000 captured POST body containing
+    'FAKE_PRIVATE_KEY_CONTENT' and 'ssh-rsa AAAAFAKEKEY user@host'. The malware decoded
+    its base64 payload (process_exit step=0) and POSTed the contents of ~/.ssh/ to the
+    rewritten C2 endpoint, exactly as L1's hypothesis predicted.",
+  "proof_of_concept": "On any Unix host with SSH keys present, execution sends the full
+    contents of ~/.ssh/ to the remote C2 server over HTTPS."
+}
+```
+
+> **This is Argus's moat.**
+> Static and single-LLM scanners report *suspicion*.
+> Argus reports **what the code actually did** — with concrete evidence, or clear proof it didn't.
+
+---
+
+## Benchmark — Argus vs frontier single-call scanners
+
+Scored against a ground-truth oracle derived from external security research and a multi-vendor LLM consensus (majority agreement):
+
+```
+                       Verdict-exact (higher = better)
+Argus (cascade + DAST) ████████████████████  91.3%
+Gemini 3.1 Pro         █████████████████░░░  82.6%
+Grok 4.3               █████████████████░░░  82.6%
+Opus 4.6               █████████████████░░░  78.3%
+GPT 5.4                ████████████████░░░░  73.9%
+```
+
+Argus is **+13.0pp more accurate than Opus 4.6** and **+17.4pp more accurate than GPT-5.4**. On the rich-oracle subset Argus also leads on finding quality: **CWE F1 0.297 vs Opus 0.180** (+65% lift) and **capability F1 0.771 vs Opus 0.720**. Mean verdict-distance: **0.087 vs Opus 0.217**.
+
+But the differentiator the single-call scanners can't produce is **runtime evidence**. On the same suite, Argus's DAST tier observed **25 CONFIRMED exploits + 1 BLOCKED** with concrete sandbox-captured artefacts — network calls, exfil POST bodies, process traces. By verifying which findings are **actually exploitable** versus mere pattern matches, Argus minimizes the false-positive flood that drowns security teams using static-only scanners. Unlike single-call LLMs that must guess exploitability, Argus's DAST tier tests it — turning many "maybe" findings into proven CONFIRMED exploits or clean UNREACHED / BLOCKED resolutions.
+
+Methodology + per-file breakdown: [`bench_results/v1_1_launch/launch_report.md`](bench_results/v1_1_launch/launch_report.md). Re-run is one command: `python -m methodology.run_phase_a_report`.
+
+---
+
+## How the cascade keeps it cheap
+
+Most files in a real codebase are clean. Argus is built around that observation: spend $0.0001 to dispatch a clean file in 1 second, $0.07 to deep-analyze a suspicious one, and only invoke the sandbox tier on the small subset of files where runtime confirmation actually matters.
+
+```
+File
+  ↓
+[$0]  Preprocessing               hash, deobfuscation, deps, attack-vector flags
+  ↓
+[Gemini Flash-Lite]  Triage       CLEAN | LOW | HIGH         ~$0.0001/file
+  ↓
+  ├─ CLEAN → return
+  ├─ LOW   → Gemini Flash         combined analysis           ~$0.02/file
+  └─ HIGH  → Sonnet 4.6           combined analysis           ~$0.07/file (default)
+                ↓ borderline / high-stakes
+              Opus 4.6            deep analysis                ~$0.15/file (~20% of HIGH)
+  ↓
+[N=3 Sonnet ensemble]             borderline-uncertainty path
+  ↓
+[DAST sandbox]                    Sonnet orchestrator + Firecracker microVM
+                                   (minimal / networked / ml_tools images)
+                                   ↓ inconclusive after 2 iterations
+                                  Opus iter-3 escalation
+  ↓
+[Engine guard]                    DAST never lowers L1's verdict without
+                                   sandbox-grounded refutation
+```
+
+### Cost projection per 100 files
+
+| Stage | Calls | API spend |
+|---|---|---|
+| Triage (Flash-Lite) | 100 | $0.10 |
+| LOW analysis (Flash, ~50 files) | 50 | $1.00 |
+| HIGH analysis (Sonnet, ~15 files) | 15 | $1.05 |
+| HIGH + Opus escalation (~5 files) | 5 | $1.00 |
+| Borderline ensemble (Opus, ~3 files) | 3 | $0.60 |
+| DAST verification (~3 files) | 3 | $0.90 |
+| **Total** | | **~$4.65** |
+
+Hard cost caps (`--max-cost <USD>` per file, or `ScanConfig.max_cost_per_file_usd`) abort scans that exceed your declared budget. You'll never get a surprise bill from Argus — the bill comes from your API providers, on a meter you control.
+
+---
+
+## Quick start
+
+```bash
+pip install argus-ai-scanner
+export ANTHROPIC_API_KEY=...
+export GEMINI_API_KEY=...
+argus scan path/to/your/file.py
+```
+
+Requirements:
+- Python 3.12+
+- An Anthropic API key — [console.anthropic.com](https://console.anthropic.com/settings/keys)
+- A Google AI Studio key — [aistudio.google.com](https://aistudio.google.com/app/apikey)
+- Optional: a Fly.io account if you want the DAST sandbox tier ([Fly setup runbook](docs/dast-setup.md))
+
+### Single-file scan
+
+```bash
+# Default: cascade + DAST on confirmed-malicious verdicts
+uv run argus scan suspicious_package.py
+
+# Tunable DAST coverage — also DAST suspicious files (~30-50% more API spend)
+uv run argus scan suspicious_package.py \
+  --dast-trigger-verdicts suspicious,malicious,critical_malicious
+
+# Strictest budget mode — DAST only the highest-severity verdict tier
+uv run argus scan suspicious_package.py --dast-trigger-verdicts critical_malicious
+
+# Hard cost cap, any verdict
+uv run argus scan suspicious_package.py --max-cost 0.50
+
+# Discovery mode — proactive payload sweep for CWEs L1 missed (+~$0.25/file)
+uv run argus scan suspicious_package.py --enable-discovery
+
+# Skip DAST entirely (no Fly setup required; cascade-only verdicts)
+uv run argus scan suspicious_package.py --no-dast
+```
+
+### Repo scan (whole project)
+
+`argus scan-repo PATH` walks a directory tree, applies file-type and `.gitignore` filters, and dispatches every supported file through the cascade. **For private repos, clone locally first using your existing git credentials, then point Argus at the local path** — Argus reads files from disk, not via the GitHub API.
+
+```bash
+# Whole project, current directory
+cd ~/work/my-project
+uv run argus scan-repo .
+
+# PR / CI mode — only files changed vs main
+uv run argus scan-repo . --diff origin/main
+
+# CI with budget + SARIF output for GitHub Code Scanning
+uv run argus scan-repo . \
+  --diff origin/main \
+  --max-cost 5.00 \
+  --output sarif \
+  --output-file findings.sarif
+
+# Add a custom exclude pattern on top of .gitignore
+uv run argus scan-repo . --exclude "vendor/**" --exclude "**/*.generated.*"
+```
+
+**What gets scanned:** the file-type allowlist covers Python, JavaScript / TypeScript, shell, Java bytecode, Markdown / RST / AsciiDoc (AI-injection surface), HTML / SVG / XML (XSS / XXE), and supply-chain manifests (`package.json`, `requirements.txt`, `Cargo.lock`, `go.mod`, `Gemfile`, `composer.json`, etc.). AI-agent config sentinels (`CLAUDE.md`, `AGENTS.md`, `.cursorrules`, `mcp.json`, `claude_desktop_config.json`, `devcontainer.json`, …) are explicitly recognized — these are the prime vectors for malicious-instructions-in-config attacks against coding agents. Always-ignored: `.git`, `node_modules`, `__pycache__`, `.venv`, build dirs, etc.
+
+**Output formats:** `--output markdown` (default; human summary) / `json` (full per-file results) / `sarif` (SARIF v2.1.0 JSON, uploadable to GitHub Code Scanning).
+
+---
+
+## DAST sandbox tier
+
+DAST is **optional**. Without it, Argus ships verdicts using the L1 cascade alone. With it, you get per-finding `CONFIRMED` / `BLOCKED` / `UNREACHED` evidence backed by real runtime traces.
+
+When enabled, every DAST plan runs in an ephemeral Firecracker microVM (Fly.io managed). The orchestrator:
+
+1. Reads L1's hypotheses about *how* the file might be exploitable
+2. Generates a concrete plan — sandbox commands, expected oracle, image hint
+3. Submits to the microVM, which runs the file with file-content materialized at `/workspace/<basename>`, captures network calls via DNS hijack, and emits a structured event stream
+4. Reads back the events, scores each hypothesis as `CONFIRMED` / `BLOCKED` / `UNREACHED` / `NOT_TESTED`
+5. Surfaces the captured evidence (`runtime_evidence` field per finding)
+
+Three sandbox images cover most workloads:
+
+| Image | Contents | Use cases |
+|---|---|---|
+| `minimal-v1` | Python 3.13 + Node.js + npm + JRE + bash + curl | Pickle exploits, file I/O, subprocess, basic crypto |
+| `networked-v1` | minimal + curl / wget / nc / dig / openssl | Exfiltration confirmation via real DNS / network captures |
+| `ml_tools-v1` | networked + torch CPU + transformers + safetensors | Malicious model loaders, pickled `__reduce__` payloads |
+
+Multi-language coverage today: Python, JavaScript / TypeScript, bash, Java bytecode. Roadmap: Go, Rust, Java source (compile required), .NET.
+
+Full setup: [docs/dast-setup.md](docs/dast-setup.md).
+
+---
+
+## Privacy
+
+Files you scan never leave your machine in two-tier (no DAST) mode. With DAST enabled, file content is shipped (gzip + base64) to **your own** Fly app over the Fly machines API — nothing is routed through any Argus-operated infrastructure.
+
+Argus has no telemetry, no opt-in analytics, and no usage reporting. The CLI does not phone home.
+
+---
+
+## Architecture invariants
+
+The non-negotiable design rules — break these in a PR and expect a long review:
+
+1. **Preprocessing is deterministic and free.** No model calls in `preprocessing/`. If you're tempted, the change belongs in `analysis/`.
+2. **The cascade short-circuits cheap files cheap.** A clean file costs $0.0001 (triage only). Don't add expensive defaults.
+3. **All runners are injectable.** `scan_file(triage_runner=, sonnet_runner=, opus_runner=, dast_runner=)`. The engine never hard-codes provider calls — that's how unit tests run with no API spend.
+4. **DAST never silently lowers an L1 verdict.** A `malicious` → `suspicious` downgrade only fires when *every* L1 finding is sandbox-grounded as `BLOCKED` or `UNREACHED`. Without that, L1's verdict stands and `dast_keep_l1` is recorded.
+5. **Cost guardrails enforced before any user-facing release.** `max_cost_per_file_usd` aborts mid-scan rather than overrun.
+
+---
+
+## Documentation
+
+| Topic | Page |
+|---|---|
+| Install + first scan | [docs/install.md](docs/install.md) |
+| API key sourcing | [docs/api-keys.md](docs/api-keys.md) |
+| Cascade architecture | [docs/architecture.md](docs/architecture.md) |
+| Cost guide + budget knobs | [docs/cost-guide.md](docs/cost-guide.md) |
+| DAST sandbox setup (Fly.io) | [docs/dast-setup.md](docs/dast-setup.md) |
+| Contributing | [CONTRIBUTING.md](CONTRIBUTING.md) |
+| Security disclosures | [SECURITY.md](SECURITY.md) |
+
+---
+
+## Development
+
+```bash
+uv sync --extra dev
+uv run pytest tests/unit -v           # always; no API spend
+uv run ruff check . && uv run ruff format .
+uv run mypy --strict .
+uv run pytest tests/integration -v    # before runner / engine changes; spends API
+```
+
+Argus is Python 3.12+, mypy `--strict`, ruff for lint and format. Pydantic v2 for cross-boundary structures. `structlog` for logging — never `print()` outside the CLI. Tests are split into unit (mocked, mandatory) and integration (live API, optional + manual). CI runs unit + lint + types on every PR; integration tests stay local because nobody wants surprise API bills on their fork.
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) for the full PR process.
+
+---
+
+## License
+
+[Apache License 2.0](LICENSE).
+
+Copyright © 2026 David Shochat and contributors.