dexterite/PR Guard AI

AI-powered code review — runs language-agnostic security, quality, and compliance checks against your pull requests using any OpenAI-compatible LLM.

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
api-keyAI API key (OpenAI-compatible). Store as a GitHub secret.yes
api-base-urlBase URL for the AI API (OpenAI, Azure OpenAI, Ollama, etc.)nohttps://api.openai.com/v1
modelModel identifier to use for analysisnogpt-4o
checksComma-separated list of checks to run, or "all". Built-in: code-quality, sast, secret-detection, iac-security, container-security noall
config-filePath to a pr-guard.config.yml with per-check overridesno""
custom-checks-dirPath to a directory with custom check definitions (prompt.md + config.yml)no""
full-scanScan ALL tracked files regardless of git diff (overrides diff-only)nofalse
diff-onlyAnalyze only files changed in the PR / push (true/false)notrue
severity-thresholdMinimum severity that causes a non-zero exit (info, low, medium, high, critical)nohigh
output-formatReport format: markdown, json, or sarifnomarkdown
ship-toComma-separated destinations: github-summary, file, webhook, github-pr-comment nogithub-summary
ship-webhook-urlWebhook URL for shipping results (used when ship-to includes 'webhook')no""
ship-file-pathBase file path for the report (used when ship-to includes 'file')nopr-guard-report
max-file-size-kbSkip files larger than this (KB)no100
max-context-tokensApproximate token budget per AI requestno100000
exclude-patternsComma-separated glob patterns to exclude from analysisno""
github-tokenGitHub token (for PR comments). Usually secrets.GITHUB_TOKENno""
request-delay-msMinimum delay (milliseconds) between AI API calls to avoid rate limitsno0
temperatureAI sampling temperature (0.0 = deterministic, 1.0 = creative). Lower is better for code review.no0.1
api-timeoutTimeout in seconds for each AI API call. Increase for slower models (e.g. gpt-5).no300
debugEnable verbose debug logging (git commands, file filtering, AI responses)nofalse
namedescription
findings-countTotal number of findings across all checks
critical-countNumber of critical-severity findings
report-pathPath to the generated report file (if ship-to includes 'file')
exit-code0 = pass, 1 = findings at or above threshold