destinationlisted
Install: claude install-skill ntholm86/principles-of-earned-autonomy-skills-suite
# Destination
*Say what you are starting to think the human means, before they have to say it again.*
*Memory Model role: Maintains `.trail/destination.md` — the operator-held destination that anchors all other memory.*
*Renamed from Vision in v2.0.0. The artifact filename is now `.trail/destination.md`; `.trail/vision.md` is read as a fallback if only the old name exists — see "Artifact name and fallback" below.*
This skill exists for one bottleneck the rest of the suite cannot touch: **the operator's articulation cost**. The destination is whatever the human has so far managed to write down. But the human is operating from a much richer interior model — interests, focus, ethics, hunches of their own — most of it implicit. Even they cannot extract it on demand.
Meanwhile the agent has signal the operator never deliberately gave it: what gets emphasised in conversation, what gets pushed back on, what gets re-routed, what makes the operator say "no, more like this." Today that signal evaporates between sessions. Destination keeps it.
The mechanism is not autonomous; it is conversational. The agent forms guesses, surfaces them as questions, and lets the operator confirm or correct. The cost the agent pays for guessing wrong is one polite correction. The cost the operator pays for never guessing is repeating themselves until the destination is clear.
## Governing principles
Destination enacts the same three principles as the rest of the suite, with one specific emphasis: