QWED Protocol Verification

Version updated for https://github.com/QWED-AI/qwed-verification to version v4.0.1.

• This action is used across all versions by 0 repositories.

Action Type

This is a Docker action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

QWED Protocol is a GitHub Action designed to create a model-agnostic verification layer for AI systems, including large language models (LLMs) and AI agents. Its primary function is to detect and block unverified, potentially erroneous outputs—such as hallucinations or incorrect computations—before they reach production. By offering multiple verification engines and ensuring process determinism, it provides robust safeguards for improving trust and reliability in AI-driven applications, regardless of the underlying model.

Release notes

Patch release aligning the TypeScript SDK, backend API schemas, and security guard integrations introduced in v4.0.0 Sentinel Edition. Addresses all critical findings from Sentry, CodeRabbit, SonarCloud, and CodeQL during PR [#97](https://github.com/QWED-AI/qwed-verification/pull/97) review.

🆕 New Endpoints

POST /verify/process

Glass-box reasoning process verifier with two modes:

• IRAC mode — Validates that AI reasoning traces contain all four IRAC components (Issue, Rule, Application, Conclusion) with decimal compliance scoring.
• Milestones mode — Checks completion of custom milestone lists with process rate calculation.

Python SDK

result = client.verify_process(
    trace="Issue: contract breach. Rule: Article 2. Application: ...",
    mode="irac"
)

TypeScript SDK

const result = await client.verifyProcess(reasoning, { mode: 'irac' });
console.log(result.result?.score);          // 1.0
console.log(result.result?.missing_steps);  // []

Agent Security Checks

POST /agents/{id}/verify now accepts an optional security_checks object:

Blocked requests return 403 Forbidden with a descriptive detail message.

🔒 Security Fixes

Information Disclosure — /verify/rag (Sentry + CodeQL)

• Before: Error responses included "message": str(e) — leaking internal file paths, DB credentials, and stack traces to API consumers.
• After: Exceptions are logged server-side via redact_pii(str(e)). Clients receive only "message": INTERNAL_VERIFICATION_ERROR and "verified": false.

Symbolic Precision — max_drm_rate (CodeRabbit)

• Before: RAGVerifyRequest.max_drm_rate accepted float | str — allowing implicit floating-point approximation (e.g., 0.1 + 0.2 ≠ 0.3).
• After: Accepts str only (e.g., "0", "1/10"). A field_validator enforces fractions.Fraction-compatible values with range check 0 ≤ rate ≤ 1.

Response Consistency — RAG Error Payload (CodeRabbit)

• Error responses now return "verified": false instead of "is_valid": false, matching the RAGGuard success response schema.

🛠️ SDK Changes — @qwed-ai/sdk@4.0.1

New Methods

Breaking Changes

Fixes

Type Changes (types.ts)

• VerificationType enum: added Process, RAG, Security
• VerificationResultData.risk: separated into risk (category) and risk_level (severity)
• Added ProcessVerifier fields: score, process_rate, missing_steps, missed_milestones
• Added RAGGuard fields: drm_rate, chunks_checked, mismatched_count

🧪 Test Coverage

New test file: tests/test_api_phase17_endpoints.py

📦 Published Packages

Full Changelog: [v4.0.0...v4.0.1](https://github.com/QWED-AI/qwed-verification/compare/v4.0.0...v4.0.1)

• QWED-AI
• GitHub Actions