Yukkai Rules Tool

Manage persistent rules that are injected into every Yukkai session, shaping how your AI assistant behaves across conversations.

What is it?

The Yukkai Rules Tool lets you define persistent behavioral rules that are automatically injected into every conversation with Yukkai. Unlike one-time instructions that vanish when a session ends, rules survive across sessions — meaning Yukkai will always remember to follow them.

Think of rules as standing instructions for your assistant. They're ideal for things like:

  • "Always respond in English."
  • "Never use em dashes in code comments."
  • "When I ask for a summary, keep it under 5 bullet points."

How it works

Every few messages (depending on priority), Yukkai re-injects your active rules into the conversation context. This ensures the rules stay in effect even during long sessions where earlier instructions might be forgotten.

Priority levels

Rules come in two priority tiers:

Priority Re-injection frequency Use case
High Every 10 messages Critical rules that must always be present
Normal Every 20 messages Standard behavioral preferences

High-priority rules consume more context window space because they're re-injected more frequently. Reserve them for rules that absolutely must not be forgotten.

Auto-detection

When you say things like:

  • "Remember to always…"
  • "Never do…"
  • "Rule: always use…"
  • "From now on, always…"
Yukkai automatically detects these patterns and persists them as rules — no need to invoke the tool manually.

How to use it

The Rules Tool supports five actions:

1. Add a rule

"Remember: always format dates as YYYY-MM-DD"

Yukkai detects the "remember" keyword and creates the rule automatically. You can also be explicit:

"Add a rule: always write user-facing messages in English"

2. List all rules

"What are my current rules?"

This returns all active rules with their IDs, priority level, and enabled/disabled status.

3. Update a rule

"Update my language rule to also use British English spelling"

You'll need the rule ID (obtained from the list action). Yukkai can look it up for you if you describe the rule.

4. Toggle a rule

"Temporarily disable my formatting rule"

This keeps the rule stored but stops it from being injected. You can re-enable it later without re-creating it.

5. Remove a rule

"Delete my rule about bullet points"

Permanently removes the rule. This cannot be undone.

Examples

Setting a coding preference

You: "Remember, always use Swift conventions when writing iOS code."
Yukkai: ✅ Rule added (normal priority): Always use Swift conventions when writing iOS code.

Setting a communication style

You: "From now on, keep your responses concise — no fluff."
Yukkai: ✅ Rule added (normal priority): Keep responses concise with no fluff.

Setting a critical rule (high priority)

You: "This is critical: never share my API keys in plain text."
Yukkai: ✅ Rule added (high priority): Never share API keys in plain text.

Reviewing your rules

You: "Show me all my rules"
Yukkai:
  1. [active | high] Never share API keys in plain text
  2. [active | normal] Always use Swift conventions for iOS code
  3. [disabled | normal] Keep responses concise with no fluff

Tips & best practices

  • Be specific and unambiguous. "Write good code" is too vague. "Always add doc comments to public functions" is actionable.
  • Use high priority sparingly. Each high-priority rule takes up context space more frequently. Only use it for rules that are truly critical.
  • Toggle instead of delete. If you think you might want a rule back later, disable it rather than removing it entirely.
  • Review periodically. Run a list command every so often to clean up rules that are no longer relevant.
  • One concept per rule. Instead of "Always use English and keep it short and avoid emojis," create three separate rules. This makes them easier to manage individually.
  • Watch for conflicts. Two rules that contradict each other (e.g., "be concise" vs. "always explain in detail") will cause unpredictable behavior. Resolve conflicts by removing or disabling one.