nav-profile
Maintains a persistent, git-ignored user profile so Claude carries your preferences and past corrections across sessions.
Triggers
Natural-language phrases that auto-invoke the skill:
- “save my preferences”
- “remember I like…” / “remember I prefer…”
- “update my profile” / “change my preference for…”
- “show my profile” / “what do you know about me?”
- “reset my profile” / “clear my preferences”
- After a correction is detected (auto-learn mode, silent)
What it does
- Stores preferences in
.agent/.user-profile.json(git-ignored, persists across context clears) - Tracks communication (verbosity, confirmation threshold, explanation style), technical (frameworks, code style, testing), and workflow (autonomous commits, compact threshold) preferences
- Auto-learns from corrections (“No, I meant…”, “Not X, use Y”, “Always do X”) and saves them silently with a confidence score, keeping a rolling window of the last 20
- Surfaces a summary of learned corrections every 5 captures
- Syncs corrections into the knowledge graph as memories when graph integration is enabled
- Tracks active goals and auto-archives ones not mentioned for 7 days
- Does NOT store code, file contents, conversation history, or sensitive data
Example
"Remember I prefer concise explanations and functional components"Claude maps the input to profile fields, writes them to .agent/.user-profile.json, and applies the preferences in this and future sessions.
Related
- Theory of Mind — bilateral modeling this skill enables
- nav-diagnose — consumes the profile to suggest preference updates