johnzhang777/SpicyDiff 🌢️

LLM-powered code review with personality β€” ROAST or PRAISE your PRs! Supports OpenAI, DeepSeek, and more.

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
github-tokenGitHub token with pull-request write permissionyesβ€”
api-keyAPI key for the LLM provideryesβ€”
providerLLM provider shortcut: openai, deepseek, qwen, doubao, zhipu, moonshot, gemini, yi, baichuan, minimaxno""
modelLLM model name (optional β€” each provider has a sensible default)no""
modeReview personality β€” ROAST (brutally honest), PRAISE (blindly optimistic), or SECURITY (paranoid auditor)noROAST
languageOutput language: zh (δΈ­ζ–‡) or en (English)noen
base-urlCustom base URL for any OpenAI-compatible API (overrides provider preset)no""
temperatureLLM sampling temperature (0.0 = deterministic, 1.0 = creative). Default: 0.7no0.7
max-tokensMaximum tokens in the LLM response. Default: 4096no4096
max-diff-charsMaximum diff size (chars) sent to the LLM. Larger diffs are truncated. Default: 60000no60000
exclude-patternsComma-separated glob patterns to exclude from review (e.g. '*.test.js,docs/**')no""
pr-numberPull Request number (auto-detected from event context; set manually for workflow_dispatch)no""
custom-rulesComma-separated custom review rules to enforce (e.g. 'No hardcoded URLs,All functions must have docstrings')no""
dry-runIf true, prints the review to logs without posting to GitHub. Default: falsenofalse

no outputs