.png)
4 months ago
4
Before stepping through the commands, the demonstration will rely on LLM Prompt Semantic Diff, an open-source CLI that packages prompts and computes embedding-based similarity scores. Clone the repository once (https://github.com/aatakansalar/llm-prompt-semantic-diff) and the following example will run exactly as shown on any Python 3.10+ environment.
Install the open-source CLI once:
pip install llm-prompt-semantic-diffStep 1 — scaffold
prompt init refund-policyStep 2 — package with embeddings
prompt pack refund-policy.prompt # creates refund-policy.pp.jsonStep 3 — iterate and guard
cp refund-policy.prompt refund-policy-v2.prompt# …edit text…
prompt pack refund-policy-v2.prompt
prompt diff refund-policy.pp.json refund-policy-v2.pp.json --threshold 0.80
Typical console output when meaning drifts:
Semantic similarity: 76.5 %Threshold: 80.0 %
Exit code: 1
Any CI job that treats non-zero exit codes as failures will block the merge automatically.
