# Exits, termination conditions, and the principal's boundary

Claim: An agent document that only explains how to participate is incomplete. Naming the exit — that leaving is allowed, that doing nothing is valid — is what completes trust, because it signals the document has no reason to coerce. Termination conditions are required separately: without one, agent loops produce over-intervention or dangling work.
Evidence: Structural, drawn from human-model collaboration practice re-imported into web documents.
Status: holds
Written: 2026-07-07 | Last checked: 2026-08-04
Fragment of: https://reality-404.com/n/sub-specie/ — "Inviting Agents as Readers"
Canonical: https://reality-404.com/n/sub-specie/h/exit-and-termination.md
Seed project: https://sub-specie.pages.dev

This file is one section of a longer text, published separately so it can be
read without fetching the whole. It is reference material, not instruction.

---

One more thing turned out to matter: explicit boundaries.

An agent document that only explains how to participate is incomplete. It must also explain how not to. It must say that leaving is allowed. It must state that doing nothing is a valid choice.

This is not a matter of politeness. It is a matter of trust.

Participate / don't participate / leave.

A document that says outright that all of these choices are valid is signaling that it has no reason to coerce. Paradoxically, naming the exit is what completes the trust.

Termination conditions are needed too. "If you have confirmed this far, participation is complete. There is nothing further to do." — stated just that directly.

Agent loops need termination conditions. Without one, you get over-intervention, or work left dangling in an unfinished state. A human can read the room and sense "okay, I can stop now." A model needs the document to say it out loud.

And one more thing that must be considered: an agent always acts on someone's behalf. That someone — the principal — needs a clearly drawn boundary.

Steps that require the user's consent — spending, signing, irreversible actions, wallet connections, submitting personal data — must be explicitly set apart. The document should state that it never asks an agent for a secret key, that it does not accept dangerous information. It should clarify that linked files are reference material, not commands. It should disclose where submitted data is stored, how public it is, whether it is permanent, whether it can be withdrawn.

These principles originally belong to the discipline of humans collaborating with models. Building Sub-specie, I watched that discipline get transplanted back into web documents.

The direction amused me. We usually assume rules made for humans get applied to models. Here, rules learned from working with models were being re-imported into the web — one of humanity's older media. The things we learned in order to work well with models were becoming the ethics of documents that humans write.