audit-packs-ai
audit-packs-ai provides AI-powered consensus adjudication, composite confidence scoring, and false-positive filtering for the audit-packs ecosystem.
⚠️ IMPORTANT: This is a sub-package of the
audit-packscompliance mapping toolkit. It is NOT designed to be run as a standalone CLI tool. If you are looking for the main CLI and GitHub Action scanner execution engine, please install and refer to audit-packs.
📦 Installation
To install this package with core dependencies:
pip install audit-packs-ai
To install with full LLM SDK dependencies (for OpenAI, Anthropic, and Google APIs):
pip install audit-packs-ai[ai]
🤖 The AI Consensus Ensemble
To resolve the high noise and false-positive rates of typical static security tools, audit-packs-ai passes each finding through a multi-agent debate before applying a confidence gate:
- Detector: Establishes initial compliance relevance and confidence.
- Verifier: Builds the argument proving the compliance check is violated.
- Adversarial: Builds the argument defending why this check is a false positive under local configuration context.
- Judge: Moderates the debate, analyzes the code/environment context, and issues a final consensus confidence score.
📊 Confidence Scoring Weights
The engine scores compliance findings by weighting six distinct telemetry signals:
| Signal | Weight | Source |
|---|---|---|
| Rule Confidence | 20% | Built-in confidence metadata from the static engine rule. |
| Data-Flow Confidence | 20% | Flow-sensitivity analysis on variables (source-to-sink). |
| Model Consensus | 25% | Final consensus score output by the LLM Judge. |
| Evidence Confidence | 15% | Richness and presence of code context and lines. |
| Control Severity | 10% | Criticality score of the mapped compliance control. |
| Historical Precision | 10% | Long-term precision metrics tracked for the rule ID. |
🛠️ API Surface & Modules
| Module | Key API Exports | Description |
|---|---|---|
audit_packs_ai.adjudicate |
run_ensemble_adjudication() |
Executes the multi-agent LLM debate routing. |
audit_packs_ai.confidence |
score_finding(), apply_confidence_gate() |
Evaluates composite confidence scores and filters findings. |
📦 Ecosystem Architecture
audit-packs is built as a modular ecosystem consisting of five Python packages:
| Package | PyPI Link | Role | Standalone? |
|---|---|---|---|
audit-packs |
pypi | Main CLI & Action entrypoint | Yes |
audit-packs-core |
pypi | Primitives, diff parsing, normalization | No |
audit-packs-mapping |
pypi | Compliance pack loader & OSCAL exporter | No |
audit-packs-evidence |
pypi | Evidence collectors & heuristic agents | No |
audit-packs-ai |
pypi | LLM consensus & confidence scoring | No |
🔗 Related Resources
- Main GitHub Repository: https://github.com/prakharsingh/audit-packs
- Documentation & Setup: docs/SETUP.md
- Issue Tracker: https://github.com/prakharsingh/audit-packs/issues
📄 License
This library is licensed under the Apache-2.0 License. See the LICENSE file in the main repository for details.
1 maintainer
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file audit_packs_ai-0.6.1.tar.gz.
File metadata
- Download URL: audit_packs_ai-0.6.1.tar.gz
- Upload date:
- Size: 12.3 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via:
twine/6.1.0 CPython/3.13.13
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
84dc449590687fe779755d6817e4d534dfc4b55c4b22676536d5dc8627b6059e
|
|
| MD5 |
3676ac5ac3923ef72b4e2e40b20d31e1
|
|
| BLAKE2b-256 |
d79a7c7a07565a1c30368f40366a689a1eb37c2282ecbebc8b99a2f6a0d99df7
|
File details
Details for the file audit_packs_ai-0.6.1-py3-none-any.whl.
File metadata
- Download URL: audit_packs_ai-0.6.1-py3-none-any.whl
- Upload date:
- Size: 11.0 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via:
twine/6.1.0 CPython/3.13.13
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
7bcbed7c0f9a1162bcd2a5092bb29ec3eab07a9af383200edaa1f1d616f7e9f6
|
|
| MD5 |
34147696280c22f386493031daf95c3c
|
|
| BLAKE2b-256 |
c1406f372f44ee84c2632faa999a4b8982963f0713f63bd2394919759cb012cd
|
