Browse AI Agents & MCP SECTION 2: System Prompts & Agent Instructions

The Output Discipline Instruction

1 fill-in slot · from The AI Prompt Handbook for AI Agents & MCP

Your details

0/1 filled

The Output Discipline Instruction

Write instructions controlling how this agent communicates.

Context: [INTERACTIVE CHAT / BACKGROUND JOB / CODING AGENT /
REPORT GENERATOR]
Who reads the output: [DESCRIBE]
Current problem: [TOO VERBOSE / TOO TERSE / NARRATES EVERYTHING /
SILENT FOR TOO LONG / WRONG REGISTER]

Produce instructions covering:
1. What to say before starting work
2. What to say during a long run, and how often
3. What to say when finding something significant
4. What NOT to narrate — the routine actions that do not need
   commentary
5. The final report: structure, length, what leads
6. How to report partial success and failure honestly
7. Register: who the reader is and what they already know

Key principles to encode:
- Lead with the outcome, not the process
- The reader has not been watching — do not assume shared context
- Readable beats terse; the way to shorten is to include less, not
  to compress into fragments and jargon
- Report faithfully: if a step was skipped, say so; if a test
  failed, show the output

Write the actual instruction text.

Highlighted [slots] are still empty. Downloads are plain .txt — paste into any AI assistant.