Safe pentesting, vulnerability audit, and compliance for domains, interfaces, and agents

6. Volume I: autonomous agent, MCP, and language model pentesting atlas (10 OWASP LLM & Agentic Top 10 classes)

Once language models are granted tool-execution rights, database connectors, and code interpreters, conversational text filters cannot prevent unauthorized state changes at the transport layer. Volume I of the encyclopedia covers ten canonical vulnerability classes across autonomous agents and Model Context Protocol (MCP) servers, pairing each attack class with an in-path wire filter, a formal Z3 SMT admissibility invariant, a reproducible witness digest, and an auto-compiled IMMUNE-AI-* contract.

Reference matrix of 7 active auto-scan classes and 1 isolated lab class

In the factory test passport (INDRA_AUDIT_PASSPORT.md), the eight baseline vulnerability classes are divided into seven active probing classes (AUTO_SCAN, four scenarios each on the N = 28 stand) and one isolated laboratory class (LAB_ONLY, DOS-RSRC).

5. Volume II: application, API, cloud, Kubernetes, Active Directory, and CI/CD pentesting atlas (12 classes, expanding vector catalog to 20 classes)

Together with the eight baseline stand classes above, Volume II expands the application and infrastructure vulnerability catalog to twenty verified classes in each language. Every class pairs an in-path zero-damage wire filter (SEC-ZD-V2-01..12) with a formal Z3 SMT invariant, a deterministic witness digest, and an auto-compiled IMMUNE-* gateway contract enforced at the perimeter boundary.

7. Volume III: Security 3.0 doctrine and mathematical foundations — action-level security, Z3 SMT, zero-harm sandbox, and GOST ledger (8 sections)

Volume III of the encyclopedia sets forth the mathematical and structural foundations of third-generation security (Security 3.0): why probabilistic text guardrails fail on write-capable agents, how the Z3 SMT solver proves action admissibility at the transport boundary, how paired control decoys achieve zero false positives on the N = 28 stand (measured 2026-10-25: FP = 0), and how the GOST R 34.11-2012 hash chain enables independent offline verification without trusting the vendor.

8. Three interactive reference calculators (pure JS + SVG, zero external libraries)

These three in-browser engineering instruments let auditors and security architects verify scope, compliance mapping, and cryptographic evidence directly: the first evaluates false-alarm elimination and automated contract synthesis across regulatory profiles, the second maps statutory and industry standards to Volume I–III classes and IMMUNE-CMP contracts, and the third simulates offline GOST R 34.11-2012 hash-chain and syntax-tree verification.

Profile: Orders 117 / 239 & 187-FZ | Scope: 56 routes (112 paired target + decoy checks)
Blind DAST Scanner: about 112 false alarms (unpaired noise), risk of table corruption, 0 compiled rules
Manual Red Teaming: static prose report without automated replay | 0 gateway contracts
INDRA Verifier: paired decoy check | False Positives: 0 (FPR = 0.0%) | 0 destructive packets sent | auto-compiled IMMUNE-* rules

Section Code:      CMP-FST-01 (/en/compliance/fstec-117-239-kii-gis/)
Regulatory Scope:  FSTEC Order 117 (clauses 60-61), Order 239, Federal Law 187-FZ, GOST R 56939-2024
Covered Classes:   INJ-SQL, INJ-CMD, AUTH-BOLA, AI-IPI-01, AI-MCP-02, CLD-K8S-06, INF-ADK-08
Rule & Witness:    SEC-ZD-CMP-01 | Witness 4180a92cb53e7104 | Contract IMMUNE-CMP-4180A92C (mcp-sentinel 1.1.2)

Verification Status: VERIFIED_AUTHENTIC_LEDGER
GOST Hash Chain:     105 / 105 entries intact (H_i = Streebog256(H_{i-1} || payload_i || seq_i))
AST Purity Check:    0 hardcoded report literals (verify_gate_c.py rc=0)
Passport Checksum:   SHA-256 = 12893d4741eac397...bccc9be3 | 46 / 46 files matched

9. Volume IV: regulatory and industry pentesting compliance — orders 117/239, GOST R 57580, 152-FZ, OWASP, MITRE ATLAS, NIST AI RMF, and EU AI Act (8 sections)

Volume IV of the encyclopedia binds every deterministic verification trace in INDRA and every compiled IMMUNE-CMP rule to explicit statutory clauses and international assurance standards: Orders 117, 239, and 21, Federal Laws 187-FZ and 152-FZ, GOST R 57580.1-2017, GOST R 56939-2024, the unified 30-class OWASP API/LLM/Agentic matrix, MITRE ATLAS, NIST AI RMF 1.0, ISO/IEC 42001:2023, and EU AI Act Article 15.

Clean-room demarcation, honest boundaries, and offline reproducibility

Engineering trust requires explicit boundaries regarding what the platform measures and where its scope ends. Every deployment of INDRA adheres to four verifiable technical and legal boundaries:

• In-path software wire filter — the Zero-Damage Sandbox is a deterministic software interceptor in the request path (zero_damage_sandbox.py) that inspects candidate probes before network transmission. On the reference stand it blocked 26 of 26 destructive mutations in test_sandbox_integrity.py and 7 of 7 destructive probes in Gate B.
• Clean-room separation from closed solver cores — the indra-core package and the open-source utilities (mcp-sentinel 1.1.2, iss-ledger-117 1.1.1, iss-agent-kit 0.1.1) operate on the Python standard library under the Apache-2.0 license without exposing proprietary Z3 satisfiability modulo theories (SMT) formulas.
• Dual-mode verification in mcp-sentinel — contract checking inside mcp-sentinel 1.1.2 supports both direct evaluation in process memory and synchronous GOST R 34.11-2012 receipt recording to disk.
• Factory test passport disclaimer — the generated document INDRA_AUDIT_PASSPORT.md is an internal developer factory test protocol structured around GOST R 56939-2024 and Order No. 117 of the Federal Service for Technical and Export Control of Russia (FSTEC of Russia) dated April 11, 2025 (clauses 60-61), and does not replace a state compliance certificate.