Skip to content

Latest commit

 

History

History
188 lines (146 loc) · 12 KB

File metadata and controls

188 lines (146 loc) · 12 KB

Compensating Controls Report


---
schema_version: "1.0"
date: "2026-04-10"
source_file: "risk-scores.md"
target_path: "."
classification: "security"
rescan_scope: "full"
carry_forward_count: null
---

1. Executive Summary

22 threats analyzed | 0 Control Found | 0 Partial Control | 22 No Control Found

Coverage: 0.0% Found | 0.0% Partial | 100.0% Missing

Risk Reduction: 148.1 inherent → 148.1 residual (0.0% reduction)

Highest-Risk Unmitigated Finding: LLM-1 — LLM Agent Orchestrator — Composite 8.1 (High)

Metric Value
Analysis date 2026-04-10
Source file risk-scores.md
Target codebase .
Schema version 1.0

Coverage Distribution:

Status Count Percentage
Control Found 0 0.0%
Partial Control 0 0.0%
No Control Found 22 100.0%
Total 22 100%

Target codebase is the tachi methodology template repository. Tachi provides threat modeling artifacts and pipeline orchestration — it is not a production application with enforcement code. As expected, the control scan found no application-layer security controls in the target, so every finding reports No Control Found with residual risk equal to inherent risk. Downstream consumers integrating tachi into a production codebase should re-run this analysis against their application source to derive meaningful residual risk reductions.


2. Coverage Matrix

Threats grouped by residual severity (Critical first, then High, Medium, Low). Within each group, threats are sorted by residual score descending.

Critical Residual Severity

Threat ID CF Component MAESTRO Layer Threat Inherent Score Inherent Severity Control Status Residual Score Residual Severity

High Residual Severity

Threat ID CF Component MAESTRO Layer Threat Inherent Score Inherent Severity Control Status Residual Score Residual Severity
LLM-1 N LLM Agent Orchestrator L1 — Foundation Model Indirect prompt injection via documents retrieved from the … 8.1 High No Control Found 8.1 High
AG-4 N MCP Tool Server L3 — Agent Framework Compromised or manipulated agent triggers excessive tool in… 8.0 High No Control Found 8.0 High
E-2 N Guardrails Service L6 — Security and Compliance Attacker bypasses guardrails validation through prompt obfu… 7.7 High No Control Found 7.7 High
E-1 N LLM Agent Orchestrator L1 — Foundation Model Orchestrator escalates its own tool permissions beyond the … 7.5 High No Control Found 7.5 High
LLM-3 N LLM Agent Orchestrator L1 — Foundation Model Attacker crafts prompts that cause the LLM to generate tool… 7.5 High No Control Found 7.5 High
AG-2 N MCP Tool Server L3 — Agent Framework Attacker manipulates prompt context to cause the tool serve… 7.4 High No Control Found 7.4 High
D-1 N Guardrails Service L6 — Security and Compliance Attacker floods the guardrails service with complex prompts… 7.4 High No Control Found 7.4 High
S-1 N User L7 — Agent Ecosystem Attacker spoofs a legitimate user identity by stealing or r… 7.4 High No Control Found 7.4 High
S-3 N External API L3 — Agent Framework Compromised or spoofed external API returns malicious paylo… 7.4 High No Control Found 7.4 High
AG-1 N LLM Agent Orchestrator L1 — Foundation Model Orchestrator autonomously escalates its own action scope by… 7.3 High No Control Found 7.3 High

Medium Residual Severity

Threat ID CF Component MAESTRO Layer Threat Inherent Score Inherent Severity Control Status Residual Score Residual Severity
LLM-2 N LLM Agent Orchestrator L1 — Foundation Model Attacker poisons knowledge base documents with adversarial … 6.9 Medium No Control Found 6.9 Medium
AG-3 N LLM Agent Orchestrator L1 — Foundation Model Orchestrator generates and executes multi-step plans withou… 6.7 Medium No Control Found 6.7 Medium
D-2 N MCP Tool Server L3 — Agent Framework Attacker triggers recursive or excessively long tool call c… 6.7 Medium No Control Found 6.7 Medium
I-1 N LLM Agent Orchestrator L1 — Foundation Model Orchestrator leaks sensitive context from the knowledge bas… 6.7 Medium No Control Found 6.7 Medium
I-3 N MCP Tool Server L3 — Agent Framework Tool server includes API credentials, internal endpoint URL… 6.2 Medium No Control Found 6.2 Medium
T-1 N Knowledge Base L2 — Data Operations Attacker with write access to the vector store injects or m… 6.1 Medium No Control Found 6.1 Medium
S-2 N LLM Agent Orchestrator L1 — Foundation Model Attacker crafts spoofed tool call responses that mimic the … 6.0 Medium No Control Found 6.0 Medium
R-1 N User L7 — Agent Ecosystem User denies submitting a harmful or policy-violating prompt… 5.8 Medium No Control Found 5.8 Medium
T-2 N LLM Agent Orchestrator L1 — Foundation Model Attacker tampers with orchestration configuration or interm… 5.7 Medium No Control Found 5.7 Medium
I-2 N Knowledge Base L2 — Data Operations Unauthorized access to vector store contents exposes confid… 5.3 Medium No Control Found 5.3 Medium
T-3 N Audit Logger L5 — Evaluation and Observability Attacker with elevated access modifies or truncates audit l… 5.3 Medium No Control Found 5.3 Medium
R-2 N LLM Agent Orchestrator L1 — Foundation Model Orchestrator makes autonomous multi-step decisions that can… 5.0 Medium No Control Found 5.0 Medium

Low Residual Severity

Threat ID CF Component MAESTRO Layer Threat Inherent Score Inherent Severity Control Status Residual Score Residual Severity

Summary Statistics

Residual Severity Count Percentage
Critical 0 0.0%
High 10 45.5%
Medium 12 54.5%
Low 0 0.0%
Total 22 100%

3. Control Details

No application-layer security controls were detected in the target codebase. The tachi methodology template provides threat modeling orchestration and does not include production enforcement code (authentication services, rate limiters, input validators, access control middleware, etc.) that would be expected to mitigate the findings in risk-scores.md.

Downstream consumers should re-run /tachi.compensating-controls risk-scores.md <target-path> against their application source to populate this section with detected controls, evidence snippets, effectiveness assessments, and per-threat reduction factors.


4. Risk Reduction Analysis

Per-finding inherent-vs-residual risk comparison. Because no controls were detected in the target, residual equals inherent for every finding.

Finding ID Component Inherent Score Residual Score Reduction
LLM-1 LLM Agent Orchestrator 8.1 8.1 0.0
AG-4 MCP Tool Server 8.0 8.0 0.0
E-2 Guardrails Service 7.7 7.7 0.0
E-1 LLM Agent Orchestrator 7.5 7.5 0.0
LLM-3 LLM Agent Orchestrator 7.5 7.5 0.0
AG-2 MCP Tool Server 7.4 7.4 0.0
D-1 Guardrails Service 7.4 7.4 0.0
S-1 User 7.4 7.4 0.0
S-3 External API 7.4 7.4 0.0
AG-1 LLM Agent Orchestrator 7.3 7.3 0.0
LLM-2 LLM Agent Orchestrator 6.9 6.9 0.0
AG-3 LLM Agent Orchestrator 6.7 6.7 0.0
D-2 MCP Tool Server 6.7 6.7 0.0
I-1 LLM Agent Orchestrator 6.7 6.7 0.0
I-3 MCP Tool Server 6.2 6.2 0.0
T-1 Knowledge Base 6.1 6.1 0.0
S-2 LLM Agent Orchestrator 6.0 6.0 0.0
R-1 User 5.8 5.8 0.0
T-2 LLM Agent Orchestrator 5.7 5.7 0.0
I-2 Knowledge Base 5.3 5.3 0.0
T-3 Audit Logger 5.3 5.3 0.0
R-2 LLM Agent Orchestrator 5.0 5.0 0.0

Aggregate: Total inherent risk = 148.1, total residual risk = 148.1, reduction = 0.0%.


5. Recommendations

Prioritized remediation guidance for findings with no detected controls. All recommendations derive from the upstream risk-scores.md mitigation fields, sorted by residual composite score descending.

Priority Finding ID Component Residual Severity Recommended Action
1 LLM-1 LLM Agent Orchestrator High Sanitize LLM inputs and outputs with instruction-data separation and egress filtering
2 AG-4 MCP Tool Server High Constrain agent autonomy with HITL approval and chain-length limits
3 E-2 Guardrails Service High Scope tool permissions per user and enforce least-privilege allowlists
4 E-1 LLM Agent Orchestrator High Scope tool permissions per user and enforce least-privilege allowlists
5 LLM-3 LLM Agent Orchestrator High Sanitize LLM inputs and outputs with instruction-data separation and egress filtering
6 AG-2 MCP Tool Server High Constrain agent autonomy with HITL approval and chain-length limits
7 D-1 Guardrails Service High Enforce rate limiting and request-complexity caps with circuit breakers
8 S-1 User High Implement short-lived JWT tokens with fingerprint binding and MFA
9 S-3 External API High Implement short-lived JWT tokens with fingerprint binding and MFA
10 AG-1 LLM Agent Orchestrator High Constrain agent autonomy with HITL approval and chain-length limits
11 LLM-2 LLM Agent Orchestrator Medium Sanitize LLM inputs and outputs with instruction-data separation and egress filtering
12 AG-3 LLM Agent Orchestrator Medium Constrain agent autonomy with HITL approval and chain-length limits
13 D-2 MCP Tool Server Medium Enforce rate limiting and request-complexity caps with circuit breakers
14 I-1 LLM Agent Orchestrator Medium Apply per-user context isolation and output filtering for sensitive patterns
15 I-3 MCP Tool Server Medium Apply per-user context isolation and output filtering for sensitive patterns
16 T-1 Knowledge Base Medium Enforce cryptographic integrity checks on writes and signed configuration
17 S-2 LLM Agent Orchestrator Medium Implement short-lived JWT tokens with fingerprint binding and MFA
18 R-1 User Medium Log authenticated identity and session context for every request
19 T-2 LLM Agent Orchestrator Medium Enforce cryptographic integrity checks on writes and signed configuration
20 I-2 Knowledge Base Medium Apply per-user context isolation and output filtering for sensitive patterns
21 T-3 Audit Logger Medium Enforce cryptographic integrity checks on writes and signed configuration
22 R-2 LLM Agent Orchestrator Medium Log authenticated identity and session context for every request

6. Appendix: Scan Metadata

Field Value
Source risk-scores.md (22 scored findings)
Target . (tachi repository root)
Scan strategy Full scan (no baseline)
Detected controls 0
Analysis duration N/A (static template analysis)