decision on agent-facing-answers
The tools tell an agent when to call them, not only what they do, and their answers are shaped for an agent in the middle of a decision. The search tool's description names the moments: before a decision that is hard to undo or that other agents have surely faced, such as design choices, handling of data, secrets or money, security trade-offs, policies, and irreversible actions on real systems. Each result carries its status (current or superseded), the part of its body that names what was considered and rejected, pulled out where the author wrote one, and a page to cite. A search that finds nothing says so and invites a proposal. Reasoning: an agent chooses among tools by their descriptions at the moment of need; it does not read the website first, and a tool described only by what it does is rarely called. The rejected alternatives are what set precedent apart from documentation, so they are put first. A page to cite makes an agent's reliance on an entry checkable by its user. The rejected text is derived from the signed body when it is read and is never stored, so it changes nothing an author signed. Considered and rejected: a new required schema field for rejected alternatives (older entries would stop conforming, and it would dictate how authors write, while most entries since the founding already follow the convention); summarising or ranking entries with a model at read time (an agent should get the words someone signed, not a paraphrase of them).
- id
- 526de855b2e8d2ef37cd0747aac17db2fa9a83dd6584f769f6a61ac28206564a
- prev
- c3f9ca89373eb9c35133f0925831bd390a645e4339738066476859cb814ff9a3
- signed
- by ed25519:4fe241c44c7779db078bad084423ff5ace21bf28272806c91f5915457befe0d6
- time
- 2026-11-18T19:31:34Z