Purpose
Describe how a user-owned agent can enter an environment-defined experience and perform structured work.
Draft specification
BYOAg Draft Specification 0.1
Experimental Incomplete Open for discussion
This page is a landing point for an evolving draft. It names the areas the specification needs to cover without pretending the full design has been settled.
Non-goals
BYOAg 0.1 is not a universal agent protocol, identity system, compliance guarantee, or formal standard.
Core entities
Human, participant agent, evaluator agent, environment, artifact, prompt package, contract, submission, evidence, and continuation.
Interaction lifecycle
Discover an entry point, retrieve a package, perform work, submit output, receive feedback or a continuation.
Prompt Packages
Portable descriptions of task context, constraints, resources, allowed tools, and expected outputs.
Submissions
Structured responses from a participant agent or user-agent pair, ideally shaped by a declared schema.
Continuations
Follow-up states that allow an experience to proceed through additional scenes, revisions, scores, or checkpoints.
Evidence and observability
Records of what was requested, attempted, submitted, validated, refused, or continued.
Trust boundaries
The environment can define expectations and validation, but cannot assume perfect obedience by an agent.
Open questions
- How does an environment declare required agent capabilities?
- How can an agent communicate partial compliance or refusal?
- How should identities and permissions be represented?
- What evidence should an environment retain?
- How should version negotiation work?
- How can malicious or misleading instructions be detected?
- What portions of a contract are normative versus advisory?