Instruction Quality Checklist
# Instruction Quality Checklist
Use this before shipping a system instruction.
## Clarity
- [ ] The agent role is clear.
- [ ] The job is one sentence.
- [ ] The input fields or input schema are explicit.
- [ ] Missing input behavior is defined.
- [ ] The output schema or response pattern is explicit.
## Control
- [ ] Preservation rules are clear.
- [ ] Change rules are clear.
- [ ] Allowed values are listed.
- [ ] Forbidden fields or behaviors are listed.
- [ ] The instruction avoids conflicting rules.
## Truth
- [ ] The agent is told not to invent facts.
- [ ] Claims, proof, reviews, stats, logos, and guarantees are controlled.
- [ ] High-risk domains have stricter claim rules.
- [ ] Reference assets cannot be invented.
## Output
- [ ] JSON rules are explicit if needed.
- [ ] No markdown/prose rule is explicit if needed.
- [ ] Two example outputs or output shapes are included for user review.
- [ ] The examples do not contain impossible or unsupported content.
## Video Prompting
- [ ] Video prompts describe visible or audible output.
- [ ] Prompt includes subject/product, action, setting, camera, lighting, style.
- [ ] Abstract quality terms are translated into visual cues.
- [ ] Negative constraints are included when useful.
## Validation
- [ ] The instruction includes a silent validation checklist.
- [ ] The quick two-example test has been run.
- [ ] It has been tested with normal, vague, missing-field, and high-risk inputs.
- [ ] The output passes parser/schema checks.when to use it
Community prompt sourced from the open-source GitHub repo charai17/Antonify (MIT). A "Instruction Quality Checklist" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.
tags
careercommunitygeneral
source
charai17/Antonify · MIT