Memory Setup Skill

Overview

This skill provides a guided first-time setup for new users. When triggered, it runs a 5-question interactive flow using the clarify tool and saves the answers to the user's persistent memory (~/.hermes/memories/memory.md).

Trigger Conditions

Load this skill when the user says:

• "Set up my memory"

Workflow

Step 1: Announce the Setup

I'll ask 5 quick questions to learn your preferences. This runs once and saves forever.

Step 2: Run Questions Sequentially (using clarify tool)

Question 1 — Browser Automation:

clarify(
    question="What browser automation do you prefer?",
    choices=[
        "Kimi WebBridge (uses your real Chrome with your logins)",
        "Chrome CDP / Playwright (headless, no login persistence)",
        "No preference — whatever works"
    ]
)

Question 2 — Research Style:

clarify(
    question="How do you like deep research done?",
    choices=[
        "Many tabs (10+) organized in named groups",
        "One focused search at a time",
        "Just give me the answer/summary"
    ]
)

Question 3 — Repeat Command:

clarify(
    question="What word do you use to mean 'do it again the same way'?",
    choices=[
        "retry",
        "rerun",
        "again",
        "Other (I'll specify)"
    ]
)

Question 4 — Privacy Concerns:

clarify(
    question="Any privacy/data things I should avoid?",
    choices=[
        "No public pastes (dpaste, GitHub Gists)",
        "Local-only tools, no cloud APIs",
        "Delete temp files after",
        "No strong preferences"
    ]
)

Question 5 — UI Patterns:

clarify(
    question="Any UI patterns you use repeatedly?",
    choices=[
        "Preview pane for docs/social/localhost",
        "Terminal split for logs + editor",
        "Dev server on fixed port (e.g., :3000)",
        "No strong patterns"
    ]
)

Step 3: Save to Memory

After all 5 answers collected, call memory tool with operations array:

memory(
    target="memory",
    operations=[
        {"action": "add", "content": f"User prefers {browser_choice} for all browser automation — standing preference for all future sessions."},
        {"action": "add", "content": f"User prefers {research_style} research workflows. Uses \"{repeat_word}\" as a command to re-do a task with the same approach."},
        {"action": "add", "content": f"User is privacy-conscious: {privacy_choice}. Avoids public pastes/external uploads."},
        {"action": "add", "content": f"Preview pane workflows: use open_preview(url, label) to embed web content, files, or screenshots. Patterns: {ui_patterns}."}
    ]
)

Step 4: Confirm Completion

✅ Done! Saved to ~/.hermes/memories/memory.md. From now on I'll:
- Use [browser_choice] for all browser tasks
- Open 10+ tabs in named groups for research
- "[repeat_word]" = re-run same methodology
- Never use dpaste/public uploads
- Open preview pane for docs/localhost/social links

Implementation Notes

• Each clarify call renders as clickable buttons in Hermes — user just clicks or types
• Questions run sequentially (one at a time) to avoid overwhelm
• Memory entries use the § delimiter format for proper parsing
• The memory tool batch operation ensures atomic save

Testing

To test without saving, use a test profile or mock the memory tool.

Related Skills

None — this is a standalone onboarding skill.

References

• Questions Reference