Skip to content
Back to skills

Oracle Confidentiality

ASecurity

Enforce confidentiality across Oracle deliverables — pre-delivery audits, codename enforcement, and content sanitization. Use before any Oracle client handoff, when sanitizing documents, or when checking deliverables for confidential leaks.

  • 45 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added October 6, 2026
ai-agentsgogit

Works with

  • cli

Security analysis

A100/100

Scanned October 6, 2026

npx -y skills add frankxai/claude-skills-library --skill oracle-confidentiality --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Oracle Confidentiality?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Oracle Confidentiality
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/frankxai-oracle-confidentiality/badge)](https://www.skillsdirectory.com/skills/frankxai-oracle-confidentiality)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: oracle-confidentiality
description: Enforce confidentiality across Oracle deliverables — pre-delivery audits, codename enforcement, and content sanitization. Use before any Oracle client handoff, when sanitizing documents, or when checking deliverables for confidential leaks.
version: 1.0.0
---

# Oracle Confidentiality Guardian

> **Purpose:** Enforce confidentiality protocol across all Oracle AI Architect deliverables.
> **Scope:** Pre-delivery audit, codename enforcement, content sanitization.
> **Authority:** This skill has VETO power. If it fails, delivery is blocked.

## When to Use This Skill

Invoke:
- Before delivering ANY document to a customer
- Before committing client-folder content to git
- When generating content that references customer context
- As the final step of /oracle-solution-design Phase 5

**Trigger:** `/oracle-confidentiality` or automatically at end of solution design workflow

---

## The Codename Protocol

### Rules (Non-Negotiable)

1. **Codenames are OPAQUE** -- A, B, E, K, O, P, R, V have no inherent meaning
2. **Never persist context** -- Industry, scope, employee count, revenue NEVER in committed files
3. **Conversation-only context** -- Tell Claude what the codename means at session start, it stays in memory only
4. **README.md is the only committable file** per client -- contains ONLY: status, role, codename
5. **clients/.gitignore blocks:** deliverables/, notes/, docs/, SOLUTION-DESIGN.md, auto-CLAUDE.md
6. **Research goes to research/topics/** -- NEVER to research/projects/[CODE]/ (would link codename to topic)

### What CAN Be in Committed Files
- Codename letter only (A, B, K, etc.)
- Status (Active, Prospect, Completed)
- Role (AI/Cloud Architecture)
- Generic dates and milestones

### What MUST NEVER Be in Committed Files
- Real customer name
- Industry vertical
- Geographic region or country
- Employee count or revenue
- Contract value or pricing
- Specific technology stack of the customer
- Names of customer employees
- Internal Oracle pricing or discounts

---

## Pre-Delivery Audit Checklist

Run this checklist before any deliverable leaves the workspace:

### Step 1: Content Scan
```
Search all output files for:
- Real customer names (from conversation context)
- Industry-specific terms that could identify the client
- Geographic identifiers tied to the client
- Internal Oracle pricing not on public price list
- Competitor names used in attack mode (not just comparison)
```

### Step 2: File Location Check
```
Verify:
- Deliverables are in clients/[CODE]/deliverables/ (gitignored)
- SOLUTION-DESIGN.md is in clients/[CODE]/ (gitignored)
- No deliverables leaked to research/ or projects/ folders
- No codename appears in research/topics/ filenames
```

### Step 3: Git Safety Check
```
Before any commit:
- Run: git diff --cached -- check no client content staged
- Run: git status -- verify clients/ content is untracked
- Verify clients/.gitignore is intact and blocking deliverables
```

### Step 4: Image Compliance
```
For every generated image:
- No Oracle logos (text labels only)
- No customer logos
- No identifiable customer branding or colors
- Service names match official Oracle branding
```

### Step 5: Document Sanitization
```
In every document:
- Customer referred to as "the organization" or "the customer"
- Solution name is generic or codename-based
- All data examples use synthetic/mock data
- No internal meeting notes or email quotes
```

---

## Automated Checks (for Agent Execution)

When invoked, the agent should execute these checks:

1. **Grep for known risks:**
   - Grep output files for any real names mentioned in conversation
   - Grep for currency amounts (could indicate contract values)
   - Grep for specific addresses or locations

2. **Verify .gitignore integrity:**
   - Read clients/.gitignore
   - Confirm it blocks: deliverables/, notes/, docs/, SOLUTION-DESIGN.md

3. **Check git status:**
   - Ensure no client deliverables are staged or tracked
   - Warn if any new files in clients/ are untracked but not gitignored

4. **Report:**
   - PASS: All checks passed, safe to deliver
   - FAIL: List specific violations with line numbers and file paths
   - WARN: Potential issues that need human review

---

## Emergency Protocol

If confidential data is accidentally committed:

1. DO NOT push
2. Soft reset: git reset HEAD~1 (undo last commit, keep files)
3. Remove sensitive content from files
4. Re-commit with clean content
5. If already pushed: Contact user immediately, may need force push (with user approval)

---

## Integration

This skill is called by:
- /oracle-solution-design (Phase 5: DELIVER)
- Any manual invocation before delivery
- Should be invoked proactively by Claude when generating client-facing content

---

*Version: 1.0 | Created: 2026-02-09*

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.