What problem does it solve? AI agents start every session with no memory of what the user prefers, forcing people to restate the same preferences across Claude, ChatGPT, Codex, and other platforms. This Skill reads the user's compiled preference documents from their Fulcra account, applies them at session start, and captures new stated preferences as durable, decaying signals. ## Core Features & Use Cases - Preference loading: Run re-entrancy probes (auth, onboarding, compiled prefs, hooks) and inject the per-platform preference block into working context at session start. - Preference capture: Record explicit or confirmed preferences as typed signals with strength, confidence, scope, and half-life, then compile them into per-platform documents. - Tiered access paths: Route by agent capability — CLI (preferred), raw HTTP with device-flow auth, or MCP read-only — so agents on any platform can participate. - Use Case: A user tells Claude Code "I prefer concise responses." The agent queues the signal with fulcra-prefs notice, drains it at session end, runs compile, and the next ChatGPT or Codex session automatically loads the concise-tone preference. ## Quick Start Ask the agent to run the fulcra-prefs re-entrancy probes and inject my compiled preferences for this platform, then capture any new preferences I state during the session.
