| name | compliance |
| description | Evidence-led regulatory compliance playbook for GDPR, HIPAA, SOC 2, PCI-DSS, and ISO 27001 with auditable outputs. |
| allowed-tools | Read, Write, Edit, Bash, Glob, Grep, Task, TodoWrite |
| model | sonnet |
| x-version | 3.2.0 |
| x-category | security |
| x-vcl-compliance | v3.1.1 |
| x-cognitive-frames | HON, MOR, COM, CLS, EVD, ASP, SPC |
Purpose & Positioning
Deliver certification-ready compliance assessments and documentation across GDPR, HIPAA, SOC 2, PCI-DSS, and ISO 27001. The skill mirrors skill-forge structure-first rules (SKILL + examples/tests/resources) and prompt-architect clarity (explicit constraints, confidence ceilings).
When to Engage / When to Redirect
- Use when: preparing for audits, mapping controls to evidence, building remediation plans, or running continuous compliance monitoring.
- Redirect when: task is general security triage (route to
security), sandbox/network setup (usesandbox-configuratorornetwork-security-setup), or unauthorized third-party reviews.
Pre-Flight Guardrails
- Operate only with explicit authorization and documented scope.
- Protect PII/PHI with encryption in transit/at rest and least privilege.
- Never fabricate evidence; every claim requires timestamped proof.
- Work in isolated environments; avoid testing on production.
- Respect confidence ceilings: inference/report ≤0.70, research 0.85, observation/definition 0.95.
Prompt Architecture Overlay
- Extract constraints into HARD/SOFT/INFERRED with sources; confirm inferred items.
- Run two refinement passes: structure (coverage/clarity) then epistemic (evidence/confidence).
- Keep outputs in English with explicit confidence ceiling notation.
SOP (Compliance Delivery Loop)
- Scoping
- Identify frameworks in-scope, data types (PII/PHI/PCI), jurisdictions, and system boundaries.
- Confirm objectives (certification, readiness check, remediation plan).
- Control Inventory
- Map requirements to existing controls and owners.
- Capture artifacts to collect (policies, configs, logs, screenshots, pen-test reports).
- Evidence Collection
- Run automated checks (scanner outputs, config exports) plus manual sampling (≥20%).
- Tag artifacts with source, timestamp, environment, and reviewer.
- Gap Analysis
- Rate findings (critical/high/medium/low) with risk rationale and regulatory citation.
- Draft remediation tasks with owners and due dates.
- Validation & COV
- Verify fixes via retest; cross-validate via second method (tool + manual).
- Run adversarial checks for edge cases (multi-tenant data paths, logging gaps).
- Delivery
- Produce an audit-ready packet: control matrix, evidence log, remediation tracker, and executive summary.
- Store outputs under
skills/security/compliance/{project}/{timestamp}with MCP tags (WHO=compliance-{session},WHY=skill-execution).
Deliverables
- Control/evidence matrix with framework mapping.
- Findings report (severity + proof), remediation backlog, and retest status.
- Audit artifacts bundle (policies, configs, screenshots/logs with hashes).
- Executive summary with residual risk and next steps.
Quality Gates
- Structure-first: SKILL.md present; README/examples/tests/resources recommended and logged if missing.
- Evidence completeness ≥90% for scoped controls; no critical gaps open at delivery.
- Dual validation on critical/high findings; explicit confidence ceiling on every claim.
- MCP logging applied; completion checklist signed (scope, evidence, validation, delivery).
Anti-Patterns to Avoid
- Scanning or testing without authorization.
- Mixing production data into test artifacts.
- Confidence inflation or missing ceilings.
- Unmapped controls (no requirement → control → evidence chain).
Output Format
- Summary of scope and frameworks.
- Constraints table (HARD/SOFT/INFERRED + confirmations).
- Control/evidence highlights and remediation queue.
- Confidence line:
Confidence: X.XX (ceiling: TYPE Y.YY) - reason.
Confidence: 0.72 (ceiling: inference 0.70) - SOP rewritten with skill-forge structure, prompt-architect constraint handling, and compliance-specific guardrails.