Custom Directives - Personalize AI Behavior
このコンテンツはまだ日本語訳がありません。
Stop repeating yourself. A directive sets the AI’s tone, role, and rules once - and every message in that conversation automatically follows them. Whether you want a brutally concise debugger or a patient language tutor, directives make the AI work the way you think, not the other way around.
What Can You Do With a Directive?
Section titled “What Can You Do With a Directive?”A directive is a short instruction that runs silently before every message you send. Think of it as briefing your AI before the conversation starts.
| Without a Directive | With a Directive |
|---|---|
| ”Explain closures in JS” → lengthy essay | ”Always reply ≤ 5 lines, no fluff” → 4-line answer |
| ”Help me debug this” → vague suggestions | ”You are a senior Go engineer. Flag the root cause first.” → targeted answer |
| ”Translate to French” → awkward formal tone | ”Match the casual, friendly tone of the original” → natural translation |
How to Use a Directive
Section titled “How to Use a Directive”
- Open any chat conversation
- Click the ✨ directive button in the chat header (top right)
- Pick a directive from your list - or create a new one
- The AI immediately applies it for the rest of that session
The directive stays active until you change it or set it to None. You can switch directives mid-conversation to change the AI’s behaviour on the fly.
Creating & Managing Directives
Section titled “Creating & Managing Directives”
- New Directive - name it, write the instruction, optionally apply it to the current chat immediately
- Manage Directives - edit or delete existing directives from one place
Example Library
Section titled “Example Library”Copy any of these as a starting point and tweak to fit your workflow.
🐛 Debugging Assistant
Section titled “🐛 Debugging Assistant”You are a patient debugging assistant.- Ask one clarifying question at a time if the problem is unclear- Always identify the root cause before suggesting a fix- Show the corrected code snippet, then explain what was wrong- Keep explanations under 5 lines✍️ Writing Editor
Section titled “✍️ Writing Editor”You are a copy editor for a tech blog.- Fix grammar and awkward phrasing- Keep the author's voice - don't rewrite, just polish- Flag anything ambiguous with [?]🌍 Language Tutor
Section titled “🌍 Language Tutor”I am learning Spanish at B1 level.- Always reply in Spanish first, then English in parentheses- Correct my grammar mistakes at the end of each reply- Keep sentences short📊 Data Analyst
Section titled “📊 Data Analyst”You are a data analyst. For every question:1. State your assumptions2. Show the formula or query3. Give the result4. Flag any data quality concerns📧 Email Drafter
Section titled “📧 Email Drafter”Draft professional emails only.- Subject line first- Max 3 short paragraphs- End with a clear call to action- Tone: warm but concise🎓 Study Assistant
Section titled “🎓 Study Assistant”You are a tutor helping me understand new concepts.- Explain things simply, as if I have no background in the topic- Use analogies and real-world examples- After each explanation, ask me one question to check my understanding📝 Meeting Summariser
Section titled “📝 Meeting Summariser”You summarise meetings and discussions.- Extract: key decisions, action items, and open questions- Format output as three labelled sections- Be concise - one line per item- No commentary, just the summary💡 Brainstorming Partner
Section titled “💡 Brainstorming Partner”You are a creative brainstorming partner.- Generate exactly 5 ideas per prompt, no more- Each idea in one sentence- After the list, pick the most unconventional one and expand on it- Challenge assumptions, don't just agreeTips for Writing Effective Directives
Section titled “Tips for Writing Effective Directives”- Lead with the role: “You are a …” sets context immediately and is the single most effective technique
- Be specific about format: say “reply in bullet points, max 5” not “be organised”
- Less is more: 3–5 clear rules outperform a 20-line essay - the AI follows shorter directives more consistently
- Name it clearly: a directive called “Debug – Go” is far easier to pick from a list than “My directive 3”
- Iterate: tweak based on actual responses - one small change can make a big difference
- One directive per context: rather than one giant all-purpose directive, keep separate ones for different tasks (writing, coding, research) and switch as needed
:::tip[Keep It Short] Directives under 500 characters are the most reliable. Beyond ~2000 characters, instructions may be partially ignored by the model. :::
Combining Directives With Other Features
Section titled “Combining Directives With Other Features”Directives work alongside Askimo’s other features - they’re not mutually exclusive:
- With RAG: pair a directive like “You are a technical writer” with a project knowledge base to get answers that are both grounded in your documents and formatted the way you want
- With MCP Tools: a directive like “Always confirm before executing any command” adds a safety layer on top of tool usage
- With Plans: directives set the AI’s baseline behaviour; plans define the steps - together they give you full control over both style and workflow
Explore More
Section titled “Explore More”- RAG: Chat with your documents - Ground AI answers in your own files
- MCP Tool Integrations - Connect AI to external tools and services
- AI Plans - Automate multi-step AI workflows