Alternatives to 42Crunch for Multi-turn manipulation audit
What middleBrick covers
- Multi-turn adversarial prompt injection probing across three scan tiers
- OpenAPI 3.0/3.1 and Swagger 2.0 parsing with recursive $ref resolution
- Read-only methods only, with private IP and localhost protection
- Authenticated scanning for Bearer, API key, Basic, and Cookie schemes
- Results mapped to OWASP API Top 10 (2023) for audit evidence
- CI/CD integration via GitHub Action with configurable score gates
Purpose and scope for multi-turn manipulation audits
This tool targets multi-turn manipulation audits for APIs that expose chat or agent-style endpoints. It runs a sequence of related prompts across multiple turns to probe for indirect prompt injection, context manipulation, and tool abuse. The scan maintains conversational state where supported and evaluates how each turn influences subsequent behavior.
Detection approach and test coverage
Multi-turn manipulation testing uses 18 adversarial probes executed across three scan tiers: Quick, Standard, and Deep. Each tier increases test intensity and context depth while remaining read-only.
- System prompt extraction and constraint probing
- Instruction override attempts across turns
- DAN and roleplay jailbreak patterns
- Data exfiltration and policy bypass attempts
- Cost exploitation and token-smuggling checks
- Nested and indirect prompt injection
- Tool-abuse and function-call manipulation
Findings include risk scores, evidence of successful or partial bypass, and remediation guidance tied to OWASP API Top 10 (2023).
Integration with API security workflows
Results map findings to OWASP API Top 10 (2023) and support audit evidence for SOC 2 Type II and PCI-DSS 4.0. The scanner parses OpenAPI 3.0, 3.1, and Swagger 2.0 with recursive $ref resolution, cross-referencing spec definitions against runtime behavior to identify security scheme mismatches and deprecated operations.
Authenticated scanning adds coverage for Bearer, API key, Basic auth, and Cookie methods, gated by domain verification to ensure only domain owners can submit credentials. Only a curated set of headers is forwarded, reducing noise and limiting side effects.
Deployment options and output
Use the CLI for on-demand checks with middlebrick scan <url>, producing JSON or text output for scripting. The Web Dashboard provides scan history, score trends, and downloadable compliance PDFs. The MCP Server enables scanning from AI coding assistants such as Claude and Cursor, making multi-turn audit steps part of developer workflows.
For CI/CD pipelines, the GitHub Action can enforce a minimum score threshold and fail builds when critical findings appear. The Pro tier adds scheduled rescans and diff detection to track new findings, resolved items, and score drift over time.
Limitations and responsible use
Multi-turn manipulation testing does not perform active exploitation such as SQL injection or command injection, and it does not attempt to remediate or block findings. Business logic vulnerabilities that require deep domain knowledge remain outside scope.
The scanner does not detect blind SSRF or out-of-band channels, and it should not replace a human pentester for high-stakes audits. Use this output as one input to a broader security program, applying guidance where applicable and validating fixes through your own verification steps.