Privacy Shield

Manages sensitive data access by marking resources with privacy levels and enforcing share, export, and usage restrictions based on a centralized registry.

Security Scan

The skill's code and runtime instructions are coherent with its stated purpose (a local privacy-marking and audit CLI) and do not request unrelated credentials or external installs.

Purpose & Capability

Name/description (privacy marking, enforcement, and audit) matches the included CLI script and SKILL.md instructions. No unrelated environment variables, binaries, or external services are required.

Instruction Scope

SKILL.md instructs agents to run the included script before any data-output operation; the script reads/writes local files (data/privacy-registry.json and data/privacy-audit.jsonl) under the WORKSPACE (or repo-root) — this is expected for the stated purpose but means the agent will access and modify workspace files and create audit logs.

Install Mechanism

No install spec or external downloads are present; the skill is instruction-only with a bundled Python script. No network installs or archive extraction are used.

Credentials

The skill does not require credentials or extra environment variables. It optionally respects WORKSPACE to locate registry/audit files, which is reasonable for a local file-based tool.

Persistence & Privilege

always is false and the skill does not request persistent platform privileges. It will read/write its own registry and audit files only; nothing in the code modifies other skills or system-wide settings.

Assessment

This skill appears to do what it says: a local privacy registry and CLI that checks and logs decisions. Before installing or invoking it, consider: (1) it will read/write data/privacy-registry.json and data/privacy-audit.jsonl in the agent workspace (or the directory implied by WORKSPACE) — review and set appropriate file permissions; (2) confirm you are comfortable allowing the agent to execute the included Python script (it runs locally and does not contact external endpoints); (3) review registry contents and audit logs for sensitive items before sharing them with other tools or people; (4) if you set WORKSPACE, ensure it points to the intended workspace so the registry/audit files land where you expect. No credentials or network calls were found in the code.

Privacy data marking system — unified sensitive resource management with pre-operation checks and audit logging.

Design Principles

• Conservative by default: Unmarked data should be handled with caution
• Check before act: Verify privacy level before any data operation
• Centralized control: One registry file governs all privacy rules

Privacy Levels

Level	Meaning	Example
owner_only	Owner can view/use only	Photos, personal info
private	No external disclosure, internal use OK	Model info, API keys
no_export	Must not leave the machine	Memory files, ontology
public	Free to use	Non-sensitive data

Quick Start

Mark a resource

# Mark a file or directory
python scripts/registry.py mark media/images/people/ --level owner_only --reason "Family photos"

# Mark a rule category
python scripts/registry.py mark --type rule "model_info" --level private --reason "Infrastructure"

Check permission

# Check if a resource can be shared
python scripts/registry.py check media/images/people/photo.jpg --action share

# Check privacy level only
python scripts/registry.py check media/images/people/photo.jpg

List all marks

python scripts/registry.py list
python scripts/registry.py list --level owner_only

Remove a mark

python scripts/registry.py unmark media/images/old-photo.jpg

View audit log

python scripts/registry.py audit
python scripts/registry.py audit --deny-only

Agent Integration

Before any data output operation:

1. Check registry → python scripts/registry.py check <resource> --action <action>
2. Evaluate result:
   • owner_only → Output only when owner requests
   • private → No external disclosure, internal processing OK
   • no_export → Never leave this machine
   • Unmarked → Default to caution, ask the owner

Registry File

Location: data/privacy-registry.json

{
  "version": "1.0.0",
  "rules": {
    "photos": {"level": "owner_only", "reason": "Family photos"},
    "model_info": {"level": "private", "reason": "Infrastructure"}
  },
  "resources": [
    {
      "path": "media/images/people/",
      "level": "owner_only",
      "reason": "Family photos",
      "marked_at": "2026-03-20T09:53:00+08:00"
    }
  ]
}

Features

• ✅ Path prefix + glob pattern matching
• ✅ Automatic audit logging (JSONL)
• ✅ Integration with image-manager (--privacy flag)
• ✅ CLI for mark/check/list/unmark/audit