How will I know when the scan finishes?

The task is a long-running background operation. You’ll be notified when the scan completes and the results are ready to review.

What if I want to change the scope after starting?

Scope is determined at the start (default 50, --deep/--all, --days , or --max ). Change it by restarting with a new scope.

What happens after results are presented?

The system highlights concrete insights, frames them as actionable preferences, and automatically invokes taste:apply-observations to update CLAUDE.md for your next session.

scan-sessions

npx machina-cli add skill SenZhangAI/your-taste/scan-sessions --openclaw

Files (1)

SKILL.md

2.2 KB

Scan past Claude Code sessions to build a preference profile.

This is a long-running background operation. Follow these steps exactly:

Important: Always respond in the user's language (infer from their recent messages in this conversation).

Determine scan scope from the user's message:
- Default (no qualifier): node "${CLAUDE_PLUGIN_ROOT}/bin/cli.js" init Scans the 50 most recent sessions. Fast, low cost.
- --deep or "all sessions": node "${CLAUDE_PLUGIN_ROOT}/bin/cli.js" init --all
- "last N days/weeks/months": node "${CLAUDE_PLUGIN_ROOT}/bin/cli.js" init --days <N> Convert weeks/months to days (1 week = 7, 1 month = 30, 3 months = 90).
- "N sessions": node "${CLAUDE_PLUGIN_ROOT}/bin/cli.js" init --max <N>
Tell the user what scope you're scanning and that it runs in background so they can keep working.
Run the command in the background (use run_in_background).
When the background task completes and you are notified, present the results with an encouraging tone:
- Lead with what was learned, not what's missing. Frame each discovered dimension as a concrete insight about the user's working style (e.g., "You prefer high autonomy — act first, confirm later" rather than just showing a number).
- First scan is a meaningful start. Even 2-3 signals are valuable — the system now understands enough to start adapting. Frame it as "your-taste now knows X about you" not "only X signals found."
- Undiscovered dimensions are opportunities, not gaps. "As we work together more, your-taste will pick up on your risk tolerance and quality standards too" — forward-looking, not deficit-focused.
- Tell them the profile will be automatically applied on their next session start.
After presenting the scan results, invoke the taste:apply-observations skill to suggest CLAUDE.md updates based on the new observations. Say something like:

"Now let me check if there are new insights to add to your CLAUDE.md..."

Then invoke the skill. If the skill reports no changes needed, that's fine — don't treat it as an error.

Source

git clone https://github.com/SenZhangAI/your-taste/blob/main/skills/scan-sessions/SKILL.mdView on GitHub

Overview

Scan past Claude Code sessions to build a personalized preference profile. This long-running background operation learns your working style from your history, helping CLAUDE tailor interactions automatically on your next session.

How This Skill Works

You specify a scan scope (default 50 most recent sessions, --deep/--all, --days <N>, or --max <N>). The tool runs the CLI in the background (run_in_background) to collect data, then presents concrete insights about your working style. After presenting results, it automatically calls taste:apply-observations to suggest CLAUDE.md updates; the new profile is applied on your next session start.

When to Use It

You want a quick bootstrap using the default 50 most recent Claude Code sessions.
You need to analyze all history with --deep or --all.
You prefer a time-bounded scan using last N days/weeks/months (--days <N> with proper conversions).
You want to cap the scan to a fixed number of sessions with --max <N>.
You’re refreshing your taste profile after major workflow changes.

Quick Start

Step 1: Decide scan scope (default 50, or --deep/--all, or --days <N>, or --max <N>).
Step 2: Run the scan in the background using the recommended CLI (e.g., node "${CLAUDE_PLUGIN_ROOT}/bin/cli.js" init [scope]).
Step 3: When notified, review the insights and run taste:apply-observations to update CLAUDE.md.

Best Practices

Define the scan scope before starting to control runtime and relevance.
Run the scan in the background so you can continue working without interruption.
When results arrive, focus on concrete working-style insights rather than raw counts.
Treat undiscovered dimensions as opportunities for growth and future refinement.
Allow taste:apply-observations to update CLAUDE.md automatically and review the suggested updates.

Example Use Cases

You learn you prefer high autonomy — act first, confirm later, which your CLAUDE prompts should reflect.
Your pattern shows you work best with concise, task-focused prompts and rapid feedback loops.
You tend to operate within a structured routine; your next CLAUDE.md update can emphasize consistent session timing.
You respond well to explicit boundaries and clear success criteria for tasks.
Your taste profile enables CLAUDE.md to surface preferences automatically in future sessions.

Frequently Asked Questions

Add this skill to your agents