always-on-agentlisted

# Always-On Agent ## Overview A personal agent on your own machine is four systems, not one, and they fail differently: ``` FACE what you look at reads state, never talks to the model BRAIN the agent loop event-driven, does the work PROFILE who you are, what you do private, gitignored, replaceable VAULT secrets the model never sees a value ``` **Core principle: the first pass must be incapable of doing harm, and the engine must be publishable without the profile.** Both are much harder to introduce later than to start with. ## When to use - Designing or building an agent that runs continuously on a personal machine - The laptop sleeps and the agent misses everything that happens meanwhile - Choosing how to approve actions from a phone - Deciding what the agent may do alone and what it must ask about - Keeping credentials out of prompts, logs and transcripts - The first run on a machine nobody has surveyed **Not for:** a hosted always-on service on a server you control, where none of the sleep, wake or personal-data problems apply. ## Phase zero, before anything can write ```bash python3 scripts/survey.py projects ~/ # every repo, dirty files, unpushed, age python3 scripts/survey.py secrets ~/ # which files hold keys, and which keys python3 scripts/survey.py stale ~/ # abandoned work that is not committed ``` Reads only. Never descends into a repository. Never opens a secret's **value**: it rea