Governed Scientific Agent — capability is not scientific authority

Start a governed-agent pilot See the KYE™ Agent Authority Stack™

The line: 148 skills give capability; KYE™ gives authority

A skill library answers can the agent do this? KYE™ answers the question a regulator, an IBC, or your general counsel actually asks: was this specific action authorised — by whom, under what mandate and purpose, with what evidence, and was it safe to become final? Possessing a skill is never permission to use it.

The distinction is the whole product. A benign, publisher-verified skill invoked without authority is still an unauthorised action, and a skill scanner's clean verdict is an input to the authority decision, never a grant — authenticity is not authority. KYE™ treats every skill invocation as a governed action through the KYE™ Tool & MCP Authority Register™: the register decides whether the agent may invoke the capability for this action, on this data, for this purpose.

Worked flow: an agent proposes an Opentrons run and a compound order

Here is how KYE™ governs two representative consequential actions a skilled science agent will attempt. The agent keeps its capability; KYE™ decides whether each proposed action has authority to proceed, then seals the outcome.

1. Agent proposes. Using an open lab-automation skill, the agent generates an Opentrons protocol and a downstream reagent order. Nothing has happened in the world yet — a proposal is not an action.
2. Action Admissibility™ check. KYE™ evaluates the proposed action at the boundary: actor identity, delegation chain, permitted purpose via Purpose Permission™, data scope, risk class, and required human-in-the-loop tier. The verdict is one of allow / deny / require_approval / suspend.
3. Human-in-the-loop tier resolves. A read-only database lookup may clear with a single approver; a compound order routes to two-person-with-legal review with sequence-of-concern screening; a clinical-database write starts untrusted and cannot become a system-of-record write without explicit, evidenced authority.
4. Evidence Pack™ is sealed. Whatever the verdict, KYE™ emits a signed Evidence Pack™ — actor, mandate, purpose, data scope, approval and finality — that a Replay-Proof™ trail can reconstruct from published keys alone, without trusting KYE™ or the skill vendor.
5. Finality is granted or withheld. Under Authority Finality™, the action becomes relied-upon only if authority was valid at T=0. No valid authority, no scientific finality — the agent's capability never converts into an unauthorised consequence.

Capability ≠ authority: the crosswalk KYE™ defines

KYE™ binds third-party scientific capability classes to a KYE™ action-class, a required authority level, and a per-invocation evidence requirement. These are governance bindings KYE™ defines for the capability class — not a claim that any library endorses KYE™, and not a redistribution of anyone's code. The library supplies the capability; KYE™ supplies the authority contract above it.

Each row is registry data in the same #46 authority-register shape that already governs KYE™'s own agent skills — one canonical contract, extended, never a parallel scheme. Where a scientific action touches a regulated regime, a KYE™ Authority Pack™ carries the deep framework mapping; see the KYE™ Pharma AI Authority™ packs for the clinical and drug-safety verticals KYE™ has already mapped.

Honest framing — what this page does and does not claim

KYE™ leads its category and states its scope plainly, because a regulated buyer cannot adopt a protocol that overclaims. Two honesty rules govern this page.

First, the referenced skill library is an independent, open-source project (MIT-licensed, an open Agent-Skills standard). It is named factually as an example of the capability layer KYE™ governs; no partnership, endorsement, integration, or affiliation is implied or claimed, and KYE™ redistributes none of its code. KYE™ ships no scientific skill library of its own and redistributes none of this one; a governed scientific agent is a KYE™ product built on top of such skills, and KYE™ governs whether a skilled agent's action may become final.

Second, KYE™ claims authority governance only for regimes it has actually mapped. This page does not assert enforcement of lab-safety, biosecurity, or clinical regulations that KYE™ has not deep-mapped to a control row; where a specific regime is not yet mapped, that is stated as out-of-scope rather than inflated to false coverage. Authority is the claim; regulatory enforcement follows only where a KYE™ Authority Pack™ maps the regime.

One authority spine under every scientific action

Whatever skills your agents load, they resolve to the same KYE™ runtime primitives — so you govern discovery, lab, clinical and procurement actions through one control point, not one per tool. KYE™ reuses these primitives; it never reimplements them per skill.

The result is one provable answer for every consequential scientific action a capable agent attempts: who had authority, for what purpose, with what evidence, and whether it became final under Authority Finality™.

Start a governed-agent pilot Browse KYE™ governed agents

Canonical KYE™ surfaces referenced on this page: Purpose Permission.