LakshmiN5/promptqc
ESLint for your system prompts — catch contradictions, anti-patterns, injection vulnerabilities, and token waste before production
What's novel
ESLint for your system prompts — catch contradictions, anti-patterns, injection vulnerabilities, and token waste before production
Code Analysis
13 files read · 4 roundsA rule-based linter for LLM system prompts that detects contradictions, injection vulnerabilities, token waste, and structural issues using regex patterns, sentence-transformer embeddings, and optional LLM-as-a-judge analysis.
Strengths
Clean rule-based plugin architecture with three analysis tiers (fast/full/judge), genuinely useful security checks (injection patterns, variable sandboxing via XML state machine), and thoughtful false-positive reduction in contradiction detection (opposing terms + Jaccard overlap + embedding similarity).
Weaknesses
The contradiction/redundancy detection relies on a fixed opposing-terms lexicon which limits recall, and the test suite lacks coverage for the semantic rules, LLM judge, and edge cases like multi-line XML sandboxing. The scoring system is simplistic (flat deductions) and doesn't weight issues by actual impact.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
9
Contributors
2
Files
53
Active weeks
1
Repository
Language
Python
Stars
1
Forks
1
License
MIT