Run each port's conformance kit; a port enters offers only through a pass, recorded in the definition.
nullius standard conform <other>[@version] --model M [--provider P] [--port PORT]... \
[--unsandboxed | --backend container] [--option NAME=VALUE]... [--attempts K] [--jobs DIR] [--stamp STAMP] [--from-run]| Flag | Does |
|---|---|
<other> |
KEY (its default version) or KEY@VERSION |
--model, --provider |
The kits' model; anthropic, custom, openai or openrouter |
--port PORT |
Only this port's kit (repeatable) |
--unsandboxed, --backend |
As for a run: as you on this machine, or each check in a container |
--option NAME=VALUE |
An agent's or capability's option (repeatable) |
--attempts K |
Each check K times (u0002r1 to u0002rK, one kit job); a majority passes it. Default 1 |
--jobs, --stamp |
Where kit jobs go (default jobs_standard/), and their time stamp |
--from-run |
Started by Nullius: OPENAI_API_KEY only, saved settings unread |
| Recorded | Detail |
|---|---|
proven |
Date, score, model, kit_version, skills shown, the kit job; with --attempts, "repeat": K, each count ("3/3"), each skill's "passed" |
| Skills | A benchmark's needs_skills fits only an agent whose proof shows them |
| Replaced | A pass on the current kit replaces an older proof; a fail takes the port back; a fidelity record stays |
Without --port, a kit refused for what its benchmark needs of this machine (a container engine, a program, a connection, its own installs) is said as PORT: refused: why and passed over: the other kits still run, their proofs are written, the refused port's record and offer stay as they were, and the command exits 1. When every kit is refused, nothing is written and it exits 2. A refusal of the command itself (the agent, its options, the key) and a refused port named with --port end it with 2 and nothing written.
!!! note "A new version runs nothing"
Until conform KEY@VERSION passes; bumping a pin in a file without versions asks again. The app's Prove its ports is the same.
-
The ports, their kits, their skills.
-
ACP against the agent's own command.