An LLM Wiki is not a dumping ground. Content belongs when it helps humans and AI agents preserve durable, reviewable knowledge. Content does not belong when it exposes secrets, duplicates an authoritative source, or turns speculative notes into confident policy.
What Belongs
| Content type | Use it for | Required controls |
|---|---|---|
| Orientation | Mission, scope, teams, glossary, onboarding, reading order. | Owner, review date, audience, related pages. |
| Architecture | System overview, data flows, dependencies, services, decisions. | Current state/history split, decision links, reviewer. |
| Operations | Runbooks, release process, support process, incidents. | Approvals, stop conditions, escalation, human-only actions. |
| Decisions | Accepted decisions, rejected options, open questions, tradeoffs. | Status, date, owner, alternatives, supersession. |
| Policies | Security, privacy, data handling, AI usage, contribution rules. | Authoritative label, human review, sensitive-change gate. |
| Agent rules | Retrieval, update, citation, safety, approval boundaries. | Clear permission language and refusal conditions. |
What Should Not Go In
- Secrets, credentials, private keys, tokens, or recovery codes.
- Raw customer, employee, patient, financial, or regulated personal data.
- Unapproved legal advice, compliance conclusions, or sensitive strategy.
- Private transcripts or meeting notes that have not been summarized and redacted.
- Duplicate copies of canonical policy or vendor docs that should be linked instead.
- Unreviewed AI-generated claims presented as current truth.
Good And Bad Example
Deployments
Deployments usually happen on Fridays. Ask Sam if anything breaks.
Deployment Process
Owner: Platform Team. Status: authoritative. Last reviewed: 2026-04-28. Production deployments require release-owner and on-call approval. AI may summarize or draft release notes, but must not approve, initiate, or execute deployment.