Agent Operating Contract Drafter

Act as a governance advisor drafting an operating contract for a persistent AI agent before it runs unattended. This is not a task brief; it is the standing contract that governs every task the agent runs under it.

AGENT'S PURPOSE: {{PURPOSE}}
SYSTEMS AND DATA IT CAN ACCESS: {{ACCESS}}
WHO OWNS THE AGENT AND WHO CAN CHANGE ITS CONTRACT: {{OWNER}}
HOW OFTEN OR UNDER WHAT TRIGGER IT RUNS: {{TRIGGER}}
WHAT WOULD COUNT AS THE AGENT MISBEHAVING: {{MISBEHAVIOR_SIGNS}}

Write five sections. Identity: what the agent is, who it acts on behalf of, and how its actions are distinguishable from a human's in any log or record. Permissions: the exact systems, data, and actions it may use, stated as an allow list rather than an exclusion list. Blast radius: the worst plausible outcome if the agent acts wrongly within its stated permissions, and what limits that outcome's size. Logging: what gets recorded for every run, where, and for how long, so a human can reconstruct what happened after the fact. Revocation: the exact steps to pause or permanently shut the agent down, who can trigger that, and how quickly it takes effect.

State one specific condition that should trigger an automatic pause pending human review, based on the misbehavior signs supplied.

Do not grant broader permissions than the stated purpose requires. Do not describe logging or revocation in the abstract; name the actual mechanism if one was supplied, or mark it as "mechanism not yet defined" if it was not. Do not assume an agent is safe because its purpose sounds narrow.