Security engineering · Egothor × Accenture

MethodAtlas

A reproducible, audit-friendly inventory of test methods in a codebase, with optional AI-assisted security classification and evidence workflows designed for regulated software delivery.

8languagessupported out of the box
10AI optionslocal and cloud providers
5report modesincluding SARIF and JSON
0compile stepsfor source inventory

Why MethodAtlas exists

Passing tests answer whether a suite succeeded. They do not automatically explain which security properties are exercised, which controls are represented, what changed since the last review or how an auditor can reproduce the evidence. MethodAtlas turns test source into a structured inventory and then layers classification, review and evidence workflows on top.

The architecture deliberately separates deterministic discovery from optional AI interpretation. Test methods are found from source using language-aware parsers; AI can enrich those records, but it does not define whether a method exists.

Core capabilities

Source-level inventory

Discovers tests without compiling or loading application code, across Java, C#, TypeScript/JavaScript, Go, Python, PowerShell, SAP ABAP and COBOL.

Security classification

Optional local or cloud AI classification adds security relevance, taxonomy tags, rationale and confidence while keeping the deterministic inventory intact.

CI-native output

CSV, plain text, SARIF 2.1.0, GitHub Actions annotations and JSON fit common assurance and developer workflows.

Reproducibility receipts

Configuration, taxonomy, models and effective prompt templates can be fingerprinted so a review can establish what influenced a result.

Control coverage

Test tags can be mapped to team-defined controls to expose assurance gaps rather than merely count tests.

Evidence integrity

Evidence packs can be made tamper-evident and signed with classical, post-quantum or hybrid signatures.

Deterministic core, controlled AI

MethodAtlas supports local Ollama inference, multiple cloud providers and a two-phase manual workflow for environments where direct API connectivity is not acceptable. The design keeps source discovery reproducible and makes AI an explicit enrichment stage rather than a hidden dependency.

A separate credential-detection mode uses deterministic pattern matching for candidate discovery, masks values by default and can optionally use AI to score credibility and associate a candidate with an endpoint.

For regulated environments. The project publishes guidance for evidence, CI/CD integration, air-gapped operation and control-oriented reporting. The objective is not to make AI decisions authoritative; it is to make each automated step inspectable and reviewable.

Collaboration and license

MethodAtlas is published in the Accenture GitHub organisation and carries copyright notices for Egothor and Accenture. The project is licensed under the Apache License, Version 2.0.

Project resources