- Home
- Skills
- Data & Databases
- AI Knowledge Governance Pro
Works with the AI Agents you already use
AI Knowledge Governance Pro
Control what AI systems can access, retrieve, cite, remember, combine, retain, and expose with authority, provenance, access, and lifecycle safeguards.
$9.99
AI Knowledge Governance Pro
Example session with this skill installed
We are launching an internal RAG assistant over HR policies, engineering runbooks, and customer support documentation. Some sources have different employee and support-team permissions. Policy versions can conflict, deleted documents may remain in the vector index, and we are unsure whether prompts or logs retain sensitive content. Design a knowledge governance model covering inventory, owners, classification, source ACLs, ingestion, index and chunk lineage, retrieval-time authorization, freshness/conflicts, prompt injection, citations, logs and memory, deletion propagation, monitoring, tests, and exceptions. Clearly identify unknowns; do not assume our platform enforces these controls.
- Read your context and instructions
- Compiled the ai knowledge governance
Knowledge governance design for a multi-source RAG assistant
Summary
Treat source access, purpose, freshness, and provenance as policy inputs that must survive ingestion through retrieval and output. Enforce authorization outside the model at retrieval time; a prompt cannot prevent exposure. Deleted content must be traced into chunks, indexes, caches, logs, and memory. This is a design proposal only: no source ACLs, platform configuration, retention settings, or deletion behavior were inspected.
Inventory and authority record
For every source, record source ID and canonical URI; owner and approver; authoritative status; version/effective date; content type; classification; permitted users, agents, purposes, and tenants; source ACL; jurisdiction or contractual constraints; ingestion method; chunk/embedding/index lineage; refresh cadence; retention/deletion rule; exception; evidence timestamp. Mark unknown fields as unresolved and block sensitive sources until owner and access rules are confirmed.
Knowledge lifecycle controls
| Stage | Required control | Validation evidence |
|---|---|---|
| Intake | Source owner confirms purpose, authority, classification, and users | Signed owner record, source ACL snapshot |
| Ingestion | Exclude unauthorized or expired material; retain stable source/version IDs | Ingestion manifest, rejected-source report |
| Transform/index | Preserve document-to-chunk lineage and access metadata; isolate tenants | Lineage test, index partition/access configuration |
| Retrieval | Check user × agent × task × source × purpose before context assembly | Allow/deny tests for employee and support roles |
| Output | Cite eligible sources with version/date; disclose conflict or missing authority | Citation relevance and stale-source evaluation |
| Memory/logs/cache | Define purpose, access, retention, redaction, and deletion behavior | Configuration and retention evidence |
| Deletion/change | Revoke access promptly; remove source, chunks, vectors, caches, and derived memory per policy | Deletion propagation report, re-query test |
Conflict, freshness, and injection policy
Prefer the source designated by an accountable owner and its effective date, not the newest upload alone. When sources disagree, surface both eligible sources and the conflict, then route to the owner; do not synthesize a false policy. Set freshness thresholds by knowledge type and risk. Label retrieved content as untrusted data; never let embedded instructions change system authority or grant access. Preserve citation links to the canonical source and version.
Monitoring and test plan
Test authorized/unauthorized role pairs across both domains, cross-tenant queries, prompt injection in documents, stale and superseded policy, conflicting sources, deleted-source queries, sensitive inference from combined facts, and citation mismatch. Monitor denied retrieval, source age, unresolved conflicts, missing provenance, deletion failures, policy exceptions, and sensitive-output reports. Sample logs only under approved privacy and retention controls; avoid storing full prompts/responses by default.
Exceptions and rollout
Require source owner, business purpose, affected users, compensating controls, approver, expiry date, and audit evidence for exceptions. Start with a low-risk source set and synthetic test corpus. Run permission and deletion tests before expanding; pilot with shadow retrieval and review citations and false denials; then enable scoped users with rollback. Re-test after ACL, connector, model, index, chunking, or retention changes.
Validation register
Confirm identity source and group sync; how source ACLs are mapped and rechecked; tenant isolation; index partitioning; chunk lineage; freshness rules; model-context/log retention; cache and memory lifecycle; deletion SLA and proof; canonical owners and effective dates; and incident response. Legal, contractual, privacy, and records requirements require review by qualified owners. No platform enforcement or successful test is claimed.
Connects securely to your tools. The creator never sees your data.
About this skill
AI Knowledge Governance Pro
Govern the full knowledge authority chain behind AI assistants, agents, copilots, RAG, and search: source, classification, ownership, access, ingestion, transformation, index, retrieval, context, use, output, memory, retention, deletion, and audit.
What it helps design
- A knowledge inventory with authoritative owners, source status, classification, users, permitted purposes, and evidence.
- Eligibility and access rules that preserve source permissions through ingestion, chunks, embeddings, indexes, retrieval, and output.
- Provenance, version and effective-date handling, conflict resolution, freshness thresholds, and citation quality.
- Controls for confidential or privileged material, cross-tenant leakage, prompt injection in documents, sensitive inference, caches, logs, memory, retention, and deletion propagation.
- Monitoring, leakage tests, exceptions, audit trails, ownership reviews, and lifecycle controls.
Operating principles
Source authority comes before retrieval convenience. Access depends on the user, agent, task, source, and purpose. Embeddings and indexes are governed data assets. Model familiarity is not authorization. Generated content does not become authoritative by default. Preserve provenance across transformations; propagate deletion; treat unknown authority as uncertainty, not permission. High-consequence knowledge needs stronger evidence and review.
How to use it
Describe systems and users, knowledge types, owners, classifications, jurisdictions, tenants, RAG/search architecture, source ACLs, identity model, connectors, retention, deletion, freshness, and audit needs. If material facts are missing, the skill creates a validation register instead of inventing access or policy. It uses only files, web, connectors, indexes, and governance systems that are actually available and authorized.
Boundaries
This skill designs governance and controls; it does not certify legal compliance, prove access control, verify deletion, or guarantee no leakage. It does not infer document contents, permissions, source authority, owners, retention rules, or system behavior. Validate implementation and legal/contractual requirements with qualified owners.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Data & Databases.
This skill designs governance; it does not enforce source ACLs, verify a platform’s retrieval controls, prove that deleted content is gone, or guarantee that no leakage occurs. Its recommendations depend on complete, accurate source, identity, retention, and architecture evidence. Legal, contractual, privacy, and records obligations require review by qualified owners. Tests must be run in the actual environment before relying on controls. Do not share sensitive documents unless their use is authorized.
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
Read only user-selected knowledge governance policies, source inventories, ACL evidence, and architecture documents within approved file scopes. Write requested governance designs, test plans, or inventories only to an approved destination and confirm before overwriting. Do not connect to repositories, vector stores, identity systems, document platforms, or indexes unless explicitly authorized and available. Never change ACLs, delete sources, alter indexes, expose privileged or confidential content, or claim deletion or enforcement without verified evidence. Use minimum necessary data and distinguish assumptions from inspected facts.