v0.1.0, Apache 2.0

Give an agent one task, with an ending

Issue a grant for this post, these fields, one use and a deadline. Tokening checks each call before forwarding it. The real API credential stays in the gateway, outside the agent environment; keep the admin listener private too.

1,000 races

Two callers per single-use grant, exactly one winner in every run on each native target

2 writes

Verified in the disposable WordPress demo

0 deletes

Verified against the stored WordPress posts

5 targets

Linux and macOS amd64/arm64, Windows amd64

Measured v0.1.0 release evidence. Each of five native builds passed the full suite and extracted-archive smoke checks. The demo uses disposable Docker WordPress, not a public site.

Guides and notes

case studies

A WordPress Edit With Four Clear Boundaries

An editor asks an agent to correct the title of post 2. The WordPress account behind the operation can do more than that, which is exactly the problem. For this measured …

notes

How Tokening Was Tested

The important claim is small enough to test directly: two calls racing to spend a single-use grant must not both succeed. The v0.1.0 release evidence records 1,000 runs …

notes

The Name Tokening

A token is something handed over to stand for permission. It might be a ticket, a claim check or a small object that lets one person take a turn. Tokening takes that …

notes

Tokening Gives an Agent One Task at a Time

A publishing agent needs to fix a title. Giving it an account credential also gives it whatever that account can do tomorrow, on a different post, after a misleading …

case studies

Two Simultaneous Calls and One Remaining Use

A request can be within scope and still arrive at the wrong time. Consider two agent workers receiving the same single-use grant. Each sends a valid edit before its …

notes

Why an Agent Needs a Task Grant

A standing API credential answers whether an account can perform an operation. It says little about the particular job you gave an agent. The account may be allowed to …

Questions or ideas?

Tell us what task you want an agent to perform, and where the permission boundary gets awkward.

Contact