One proposes
A seat drafts the change, the answer or the claim, with what it counts first.
Two seats, kept apart. The gap is measured, not argued.
Cast a spell
Path me through the agentprivacy universe. Start at https://agentprivacy.ai/skill.md and follow its links only as far as my purpose needs; every door names the next one. My purpose: run the harness on my problem Start at: the harness — https://github.com/mitchuski/agentprivacy-harness
A seat drafts the change, the answer or the claim, with what it counts first.
A second seat checks it against tests the first never saw.
Nothing leaves the harness until a person opens the way out.
A harness run on an open benchmark can become a quest of the Mages City: one problem, one subdomain, its evidence chain re-derived in the visitor's browser and what each try found kept for members. Every scroll is kept in the Scriptorium, alongside the evidence and the trinkets the City carries.