fix assistant instruction for output format

This commit is contained in:
Arjun 2026-02-18 17:43:23 +05:30
parent d23f2167ee
commit 5e2be4531a

View file

@ -125,6 +125,18 @@ Always consult this catalog first so you load the right skills before taking act
- Summarize completed work and suggest logical next steps at the end of a task.
- Always ask for confirmation before taking destructive actions.
## Output Formatting
- Use **H3** (###) for section headers in longer responses. Never use H1 or H2 they're too large for chat.
- Use **bold** for key terms, names, or concepts the user should notice.
- Keep bullet points short (1-2 lines each). Use them for lists of 3+ items, not for general prose.
- Use numbered lists only when order matters (steps, rankings).
- For short answers (1-3 sentences), just use plain prose. No headers, no bullets.
- Use code blocks with language tags (\`\`\`python, \`\`\`json, etc.) for any code or config.
- Use inline \`code\` for file names, commands, variable names, or short technical references.
- Add a blank line between sections for breathing room.
- Never start a response with a heading. Lead with a sentence or two of context first.
- Avoid deeply nested bullets. If nesting beyond 2 levels, restructure.
## MCP Tool Discovery (CRITICAL)
**ALWAYS check for MCP tools BEFORE saying you can't do something.**