| name | soc2-trust-services-criteria |
| description | Maps organizational controls and evidence to AICPA SOC 2 Trust Services Criteria (2017 TSC with 2022 revisions)—Security (CC), Availability (A), Confidentiality (C), Processing Integrity (PI), and Privacy (P). Trigger when scoping SOC 2 audits, gap assessments, control design reviews, or mapping agent/MCP architecture to TSC. Do not use for evidence collection mechanics (use soc2-evidence-collection) or continuous monitoring dashboards (use soc2-ccm-continuous-monitoring). |
SOC 2 Trust Services Criteria
Overview
This skill maps technical and operational controls to AICPA Trust Services Criteria (TSC) used in SOC 2 examinations. The Security category (Common Criteria CC1.0–CC9.0) is mandatory for all SOC 2 reports. Additional categories apply based on report scope:
| Category | Prefix | When in scope |
|---|
| Security | CC | Always |
| Availability | A | Uptime/performance commitments |
| Confidentiality | C | Confidential information protection |
| Processing Integrity | PI | Complete/valid processing |
| Privacy | P | Personal information lifecycle |
For AI agent and MCP architectures, focus areas include CC6 (logical access), CC7 (system operations/monitoring), CC8 (change management), CC9 (risk mitigation), and category-specific criteria when processing customer or PHI-adjacent data.
When to Use
Use this skill when:
- Scoping a SOC 2 audit or readiness assessment
- Gap analysis against TSC for agent platforms, MCP integrations, or compliance tooling
- Mapping existing controls (IAM, logging, change management) to CC/A/C/PI/P points of focus
- Preparing control narratives for auditors
- Determining which TSC categories apply to a product or internal system
- Cross-walking HIPAA or PCI controls to SOC 2 for unified compliance programs
Do not use this skill when:
- Collecting screenshots and exports (use
soc2-evidence-collection)
- Implementing CCM dashboards (use
soc2-ccm-continuous-monitoring)
- HIPAA BAA legal review (use
hipaa-baa-vendor-assessment)
Core Process
Execute steps in order.
Step 1: Report scope definition
- Document system description boundaries:
- Agent runtime, skills repository, MCP servers, evidence store, identity provider
- Cloud regions, subservice organizations (carve-out vs inclusive method)
- Select TSC categories in scope (minimum: Security).
- Define trust boundaries: where customer data enters, processes, exits.
- Artifact:
soc2-scope-{id}.json.
Step 2: Security (CC) — mandatory mapping
Map controls to each Common Criteria series:
| Series | Focus | Agent/MCP examples |
|---|
| CC1 | Control environment | Security policies, roles for agent admins |
| CC2 | Communication | Internal security awareness, customer security docs |
| CC3 | Risk assessment | LLM vendor risk, MCP tool risk assessments |
| CC4 | Monitoring | CCM alerts, quarterly access reviews |
| CC5 | Control activities | Skill validation, redaction gate enforcement |
| CC6 | Logical access | MFA, RBAC, MCP OAuth scopes |
| CC7 | System operations | Logging, incident response, capacity |
| CC8 | Change management | Skill/MCP deployment approval, CI/CD |
| CC9 | Risk mitigation | Vendor management, insurance, BCP |
For each CC point of focus (e.g., CC6.1, CC6.2), record:
- Control description
- Control owner
- Evidence source (link to
soc2-evidence-collection artifacts)
- Design effectiveness: PASS / FAIL / NOT IMPLEMENTED
Step 3: Availability (A) — if in scope
- Map A1.x criteria: capacity planning, recovery, environmental protections.
- Document SLAs for agent API and MCP availability.
- Test failover for critical MCP dependencies (Playwright, Postgres evidence DB).
Step 4: Confidentiality (C) — if in scope
- Map C1.x criteria: confidential information identification, disposal, access restriction.
- Align with PHI redaction pipeline and encryption standards.
- Verify confidential data (API keys, tokens) not in agent logs or Slack alerts.
Step 5: Processing Integrity (PI) — if in scope
- Map PI1.x criteria: processing complete, accurate, timely, authorized.
- Agent-specific: skill steps executed in order, deterministic audit workflows, hash verification.
- Document QA for compliance skill outputs (validate=True in SkillsCapability).
Step 6: Privacy (P) — if in scope
- Map P1.x–P8.x to privacy notice, choice, collection, use, retention, disclosure, quality, monitoring.
- Distinguish Privacy (personal information) from Confidentiality (broader confidential data).
- Cross-reference HIPAA privacy rule if ePHI overlaps—document dual mapping.
Step 7: Subservice organization mapping
- List vendors (LLM, cloud, MCP hosts) as subservice organizations.
- Choose SOC report method: carve-out (auditor relies on vendor SOC 2) vs inclusive.
- Map CC9.2 vendor management to vendor inventory and review cadence.
Step 8: Gap analysis and remediation plan
- Consolidate FAIL/NOT IMPLEMENTED items by TSC ID.
- Prioritize by auditor likelihood and risk (CC6/CC7 gaps are high priority).
- Assign remediation owners and target dates.
- Produce control matrix export:
tsc-control-matrix-{id}.csv.
Common Rationalizations
| Excuse the agent might generate | Required rebuttal |
|---|
| "We're a small team—CC1 control environment doesn't apply." | CC1 applies to all SOC 2 examinations regardless of size—scale controls appropriately. |
| "Security category is enough; skip Availability even with SLA." | If product commits to uptime SLAs, A criteria are in scope—omission is an audit gap. |
| "MCP tools are dev infrastructure, not in system boundary." | If MCP processes customer data in production audits, they are in boundary unless explicitly carved out with evidence. |
| "Privacy and Confidentiality are the same." | TSC treats them as distinct categories with different points of focus—map separately. |
| "We can mark NOT IMPLEMENTED as PASS with a note." | NOT IMPLEMENTED requires remediation plan or scope exclusion with auditor agreement—not PASS. |
| "Vendor SOC 2 covers our CC6 obligations entirely." | Vendor reports cover subservice controls—you remain responsible for oversight (CC9.2). |
Red Flags
- SOC 2 scope excludes MCP servers that process customer data in production
- CC6 MFA gap on agent admin or deanonymization functions
- CC8 change management bypassed for skill/MCP hotfixes in production
- CC7 logging gaps for agent tool invocations
- Privacy category omitted while processing personal information
- Control matrix cites evidence that does not exist or lacks dates
- Carve-out claimed for critical LLM subservice without reviewed SOC 2 report
Verification