AI_USAGE - mark-ik/graphshell GitHub Wiki

Acceptable AI Use Criteria

1. Only use when you can validate against standards/your planning -- a requirement when a human isn't in the loop especially. Don't freeball it, for every change you need to keep everything in sync.
2. Create a documentation schema for your planning and the AI's. Regularly make receipts. This is what keeps things in sync.
3. Make policies for AI agents accessing your repo, designate priority documents with indices of other documents. Categorize and keep your documents current with regular audits.
4. Document sprawl is to be avoided. You will need to consolidate docs periodically, create canonical docs, determine the relative authority of contradictory docs, and reconcile these contradictions.
5. Track terminology; inconsistent or misused terms lead to interpreted distinctions that can create unpredictable behavior.
6. Only use for open source: all AI-generated code needs to be auditable. But AI is an answer to a real and pressing problem: the need to process lots of data. Normal people need that capability in order to have data autonomy. Likewise, for open source projects to solve the real problems of the future, we need the appropriate tools.
7. Attestation: the contributor who has an agent act on their behalf must ultimately be responsible for that agent and explaining the agent's decisions. Undocumented, unclear, or unattested code cannot be used.
8. The programmer-in-the-loop model is strongly preferred: all issues, prs, actions should come from the contributor's account, although autonomous agents are acceptable when their originator is clear, their output is subject to review, and they are developing in accordance with a reputable standard and a spec-defined target.