Works with the AI Agents you already use
Agent Privacy Guardian — Find AI Data-Handling Risks
Map how AI workflows collect, use, expose, and retain data, then prioritize practical privacy-by-design controls and verification steps.
$9.99
Agent Privacy Guardian — Find AI Data-Handling Risks
Example session with this skill installed
Review this proposed AI lead-enrichment workflow. A sales rep uploads a CSV with business emails, names, company, and free-text notes. A model summarizes each lead; a web research connector adds public company facts; results are stored in a CRM and prompts are logged for 90 days. Unknowns: connector provider/retention, CRM access roles, deletion propagation, and whether notes can contain sensitive personal details. Build a data inventory and flow map, identify risks by likelihood/impact with evidence status, propose minimization and access controls, prompt/log retention questions, deletion tests, vendor questions, and a prioritized verification plan. Do not assume consent, lawful basis, data location, or compliance.
- Read your context and instructions
- Compiled the agent privacy guardian
Privacy review: AI lead-enrichment workflow
Executive summary
The described flow processes identifiable business-contact data and free-text notes through a model, a research connector, CRM storage, and 90-day prompt logs. The largest immediate uncertainties are the notes’ contents, connector handling, CRM access, and end-to-end deletion. Do not conclude that public information or a business email removes privacy obligations.
Data flow inventory
| Stage | Data and purpose | Known from brief | Unknown to verify |
|---|---|---|---|
| CSV upload | Names, business emails, company, notes; initiate enrichment | Sales rep uploads CSV | Collection source, notice/authorization, allowed fields, input validation |
| Model prompt | Lead context used to generate summary | Model summarizes each lead | Provider, processing region, retention/training terms, sensitive-note filtering |
| Web research | Public company facts added to record | Connector performs research | Provider, query contents, logging, source accuracy, transfer/retention |
| CRM | Enriched lead stored for sales use | Results stored in CRM | Roles, tenant boundaries, exports, downstream syncs, retention |
| Prompt logs | Prompt content retained for 90 days | 90-day period stated | Whether logs include full CSV/notes, access, purpose, deletion, backups |
Prioritized risk register
| Priority | Risk | Evidence status | Why it matters |
|---|---|---|---|
| High | Sensitive personal details in free-text notes flow to model/logs | Possible; field is unconstrained | Unnecessary special-category or private data may reach additional processors and logs |
| High | Connector/provider handling is unknown | Unknown | Lead identifiers or queries may be retained or transferred beyond intended systems |
| High | Deletion may not propagate to prompts, logs, connector, CRM, exports, or backups | Unknown | A CRM deletion alone may leave copies and derived summaries |
| Medium–High | CRM access roles and tenant boundaries are unspecified | Unknown | Enrichment can broaden who sees contact details or notes |
| Medium | Public company facts may be wrong or attached to the wrong person/company | Inferred risk | Incorrect enrichment can affect outreach decisions and trust |
| Medium | 90-day prompt logging may exceed operational need | Known duration; necessity unknown | Full prompts may duplicate source data and expand breach exposure |
Safer design and controls
Minimize at intake: define allowed columns; reject or quarantine unexpected fields; instruct reps not to enter sensitive personal details. If notes are necessary, apply a documented redaction step before model or connector transmission.
2.
Separate company research from personal data: query by company name/domain when sufficient; avoid sending names, emails, or notes to research tools unless a defined purpose requires them. Store source and retrieval date for each added fact.
3.
Limit prompts and logs: send only the minimum fields needed for summarization. Determine whether logs need full prompts; prefer structured, redacted metadata and the shortest justified retention. Restrict log access and audit reads.
4.
Scope access: define CRM roles and exports; separate tenants and teams at the service boundary; avoid making raw notes broadly visible if the summary is sufficient.
5.
Define lifecycle: map deletion across CSV staging, model/connector records, prompts, CRM records, exports, caches, backups, and derived summaries. State actual timelines only after each processor’s behavior is confirmed.
6.
Protect provenance and correction: label generated summaries and sourced facts; keep source links/timestamps; provide a correction path; do not present inferred attributes as verified facts.
Vendor and policy questions
- Which model and research providers receive each field? Are prompts retained, used for training, or visible to subprocessors?
- What regions, transfer mechanisms, access roles, incident terms, and deletion APIs apply?
- Can provider logs and derived content be deleted, including backups, and how is completion evidenced?
- What is the purpose and necessity for each field and the 90-day log period? Who approved it?
- What notices, permissions, rights handling, and retention schedule govern the contact data? Confirm legal questions with qualified counsel/privacy owners.
Verification plan
First: inspect a synthetic CSV with representative edge cases; confirm field allowlist and sensitive-note rejection/redaction.
Next: trace a single synthetic record through prompt, connector, CRM, logs, and exports; document actual recipients, fields, regions, and retention.
Then: test least-privilege roles and cross-tenant isolation. Submit a test deletion and verify every store, derived record, log, and backup behavior against provider evidence. Finally, review whether the 90-day log period is necessary and shorten it where feasible.
Unresolved: legal basis, applicable jurisdictions, provider terms, retention beyond CRM, and deletion guarantees were not supplied; no compliance conclusion is made.
Connects securely to your tools. The creator never sees your data.
About this skill
Find and reduce privacy risk across AI workflows
Agent Privacy Guardian analyzes how agents, prompts, tools, connectors, memory, retrieval, logs, analytics, exports, vendors, and human processes collect, infer, combine, transfer, retain, or expose information. It turns a workflow description into a grounded data-handling inventory and prioritized privacy engineering plan.
What it helps assess
- Data types, subjects, purpose, source, collection, processing, storage, access, sharing, and retention
- Sensitive data entering prompts, retrieval systems, memory, telemetry, logs, or exports
- Excessive connector scopes, overbroad retrieval, cross-tenant leakage, and secondary use
- Data minimization, redaction, access boundaries, consent signals, provenance, and purpose limitation
- Retention schedules, deletion paths, backups, derived data, and lifecycle ownership
- Vendor and subprocessor questions, transfers, human review, and disclosure risks
- Practical controls, verification evidence, residual risk, and follow-up questions
Evidence-based approach
The skill distinguishes supplied facts, reasonable inferences, and unknowns. It never invents data flows, consent, retention periods, subprocessors, jurisdictions, encryption, deletion guarantees, compliance status, certifications, identities, connector access, or controls. It identifies what evidence would verify a safeguard and prioritizes minimum safer designs that still meet the business goal.
Important boundaries
This is privacy engineering analysis, not legal advice or certification. It does not claim regulatory compliance. Jurisdiction-specific legal conclusions, lawful bases, notices, and rights obligations should be reviewed by qualified counsel or the responsible privacy team.
Best results
Provide a workflow diagram or steps, data fields and examples, user groups, systems, connectors, model/provider boundaries, storage and logs, access roles, purposes, retention/deletion behavior, geographic constraints, and known policies. Remove real personal data and secrets; use representative synthetic examples when possible.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
Findings depend on the completeness and accuracy of supplied data-flow evidence. Unknown provider practices, locations, retention, or controls must be verified directly. The skill does not determine legal compliance, lawful basis, or jurisdiction-specific obligations. Recommendations cannot guarantee deletion or prevent misuse without implemented and tested controls. Use synthetic or redacted examples for analysis.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Permissions
No external service is required by default. Read only documents the user supplies or authorizes. Browser/network research is optional for current jurisdictional or provider claims and must use authoritative sources when available. Do not upload real personal data to third parties for analysis without explicit authorization; prefer redacted or synthetic examples.