Add compliance baseline for ip-legal plugin

Introduce COMPLIANCE.md as a plugin-level shared guardrail covering:
- Deployment environment classification (public cloud / private / air-gapped)
- Five-tier information sensitivity framework with per-tier AI rules
- Mandatory pre-input self-assessment checklist
- Per-skill confidentiality quick-reference table
- Bright-line prohibited input rules
- Output management, labeling, and storage requirements
- Audit logging and incident reporting standards
- Privilege and work-product considerations

This file provides a jurisdiction-agnostic compliance foundation that
reinforces the design principle from CONTRIBUTING.md: "CLAUDE.md provides
plugin-level safety net." Skills read COMPLIANCE.md alongside CLAUDE.md
to determine the appropriate environment and tier for each invocation.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: oxunafufa55

ip-legal/COMPLIANCE.md

@@ -0,0 +1,114 @@
+# ip-legal Compliance Baseline
+
+> This file establishes the global confidentiality, data-handling, and output-management guardrails for the `ip-legal` plugin. Every skill and agent must read this file before execution.
+
+---
+
+## 1. Deployment Environment Classification
+
+| Tier | Definition | Plugin Suitability |
+|:---|:---|:---|
+| **Public Cloud Model** | Third-party API (Claude API, ChatGPT, etc.) — data uploaded to service provider servers | ⚠️ Public-tier information only; no Internal/Confidential/Restricted data |
+| **Private Deployment Model** | Self-hosted or private cloud (local LLM, VPC-hosted model) — data does not leave organizational control | ✅ Suitable for Public, Internal, and Confidential information |
+| **Air-Gapped / Classified Network** | Physically isolated system for national-security or trade-secret information | ⚠️ Must be approved by security officer; Top Secret / classified info prohibited from any AI system |
+
+---
+
+## 2. Information Sensitivity Tiers
+
+| Tier | Label | Examples | AI Processing Rule |
+|:---|:---|:---|:---|
+| **Public** | Public | Issued patents, published product materials, public court filings | May be submitted to any model |
+| **Internal** | Internal | Internal workflows, non-critical training materials, org charts | May be submitted to public cloud (sanitization recommended) |
+| **Confidential** | Confidential | Unpublished patent applications, invention disclosures, draft contracts, pre-litigation strategy | **Private deployment only** |
+| **Restricted** | Restricted | Core algorithms, source code, trade secrets, active litigation strategy, M&A deal terms | **Private deployment + audit logging required** |
+| **Prohibited** | Prohibited | Classified national security information, third-party information under strict NDA prohibiting AI disclosure | **Never submit to any AI system** |
+
+---
+
+## 3. Pre-Input Self-Assessment Checklist (Mandatory)
+
+Before every skill invocation, the user must confirm:
+
+- [ ] I have classified the information I am about to submit: □ Public  □ Internal  □ Confidential  □ Restricted  □ Prohibited
+- [ ] The current AI deployment environment is authorized for this tier
+- [ ] If submitting unpublished patent disclosures, core technical data, or trade secrets, I am operating in a private deployment environment
+- [ ] The input does not contain classified national security information or third-party information explicitly barred from AI processing
+- [ ] The input has been sanitized of unnecessary personal data (per applicable data protection laws)
+
+**If any item cannot be confirmed, pause and escalate to the designated security or compliance officer before proceeding.**
+
+---
+
+## 4. Per-Skill Confidentiality Quick Reference
+
+| Skill | Typical Input | Sensitivity Tier | Environment Requirement |
+|:---|:---|:---|:---|
+| `cold-start-interview` | Organizational policies, approval chains, external provider contacts | Internal / Confidential | Private deployment (recommended) |
+| `invention-intake` | Invention disclosures, technical solutions | Confidential / Restricted | Private deployment |
+| `fto-triage` | Product technical specifications, feature lists | Confidential / Restricted | Private deployment |
+| `clearance` | Proposed marks, goods/services classifications | Internal / Confidential | Private deployment (recommended) |
+| `infringement-triage` | Infringement evidence, trade secret descriptions, litigation strategy | Confidential / Restricted | Private deployment |
+| `ip-clause-review` | Contract text, license terms | Confidential / Restricted | Private deployment |
+| `oss-review` | SBOM, dependency manifests, source code snippets | Internal / Confidential | Private deployment (recommended) |
+| `portfolio` | Portfolio database, maintenance fee records | Internal / Confidential | Private deployment (recommended) |
+| `claim-chart-builder` | Patent text, litigation materials, infringement analysis | Confidential / Restricted | Private deployment |
+| `cease-desist` | Infringement facts, enforcement strategy, settlement positions | Confidential / Restricted | Private deployment |
+| `takedown` | Infringement URLs, platform account data, purchase records | Internal / Confidential | Private deployment (recommended) |
+| `matter-workspace` | Case files, correspondence, billing records | Confidential / Restricted | Private deployment |
+| `customize` | Configuration preferences, integration settings | Internal | Private deployment (recommended) |
+
+---
+
+## 5. Prohibited Inputs (Bright-Line Rules)
+
+The following information **must not** be submitted to any AI system, regardless of deployment environment:
+
+1. **Classified national security information**: Any material marked as classified, top secret, or equivalent under applicable law
+2. **Third-party confidential information**: Information received from clients, partners, or counterparties under an NDA that explicitly prohibits AI processing — unless the disclosing party has given express written consent
+3. **Core trade-secret source code**: Unpatented, unpublished core algorithm implementations whose disclosure would destroy trade-secret protection
+4. **Unredacted personal data**: Government-issued identifiers, biometric data, health records, financial account numbers, or precise geolocation — unless redacted per applicable data protection law
+5. **Undisclosed M&A / securities information**: Material non-public information that could violate insider-trading or market-abuse regulations
+6. **Attorney work product subject to litigation hold**: Information whose submission to a third-party AI service could waive privilege or work-product protection — unless counsel has cleared the specific use
+
+---
+
+## 6. Output Management
+
+1. **Mandatory disclaimer**: All outputs must carry the notice — *"This document is an AI-generated draft. It must be reviewed by a qualified legal professional before any reliance or filing."*
+2. **Output tier inheritance**: The sensitivity tier of the output is the highest tier of any input used to generate it. If Confidential inputs were used, the output is Confidential.
+3. **No direct official submission**: AI-generated application drafts, legal briefs, cease-and-desist letters, and platform complaints must not be filed with courts, patent/trademark offices, or administrative bodies without qualified human review and sign-off.
+4. **Output storage**: Confidential and Restricted outputs must be stored on organizational internal networks or encrypted storage. Do not store on public cloud note-taking services, personal devices, or unencrypted external drives.
+
+---
+
+## 7. Audit & Logging
+
+1. **Invocation logs**: Record skill name, timestamp, input summary (sanitized), output summary, and operating user for every invocation.
+2. **Retention**: Log retention periods should match the recordkeeping requirements of the underlying legal matter (e.g., patent prosecution files are typically retained for the life of the patent plus a jurisdiction-specific post-expiration period).
+3. **Access control**: Log access is limited to the security/compliance officer, the IP practice lead, and IT audit personnel.
+4. **Incident reporting**: Any suspected exposure of Restricted-tier information to an unauthorized environment, data exfiltration, or credential leak must be reported to the security officer within 24 hours.
+
+---
+
+## 8. Privilege & Work-Product Considerations
+
+1. **Attorney-client privilege**: Submitting privileged communications to an AI system may waive privilege in some jurisdictions. Confirm with supervising counsel whether the specific use is covered by the jurisdiction's privilege rules.
+2. **Patent agent privilege**: In jurisdictions that recognize patent-agent privilege (e.g., U.S. under *In re Queen's University at Kingston*), confirm the scope is limited to USPTO practice. Non-patent matters submitted to an AI system through a patent-agent role may lack privilege protection.
+3. **Work-product doctrine**: Documents prepared in anticipation of litigation retain work-product protection only if kept confidential. Assess whether AI processing in the chosen deployment environment maintains the required confidentiality.
+
+---
+
+## 9. Violation Consequences
+
+Failure to adhere to this compliance baseline may result in:
+
+- **Regulatory penalties** under applicable data protection, trade-secret, or confidentiality laws
+- **Civil liability** for damages to the organization or third parties
+- **Professional discipline** including bar or patent-office sanctions
+- **Loss of privilege or trade-secret protection** for the affected information
+- **Organizational disciplinary action** up to and including termination
+
+---
+
+*This compliance baseline is maintained by the plugin security owner. Version v1.0. Review at least annually and after any material change to the deployment environment or applicable regulatory framework.*