PR Truth Gate

PR Truth Gate is a production-grade review skill that catches high-risk AI-generated pull requests before merge. It enforces evidence-backed review decisions, trust-boundary checks, and rollback-readiness in one repeatable workflow.

Supported agents:

• Codex
• Claude
• Gemini

Why this exists

Modern teams ship faster than ever, and AI-assisted code can look correct while still introducing hidden security, reliability, or business-logic regressions. PR Truth Gate gives teams a structured gate that is strict enough for production, but fast enough for daily CI workflows.

What it solves

• Prevents merge of PRs with unsupported or risky changes
• Detects trust-boundary violations (auth, secrets, permissions, external inputs)
• Flags rollback-hostile migrations and side effects
• Requires evidence for each risk claim
• Produces a final merge verdict with confidence

Core outputs

• Risk score (0-100)
• Severity-ranked findings
• Merge verdict: APPROVE / APPROVE_WITH_GUARDS / BLOCK
• Minimal fix checklist
• Rollback and verification plan

Repository layout

• SKILL.md: Core workflow
• skill.json: Metadata
• policies/risk_matrix.md: Scoring model
• prompts/: Reusable prompts
• adapters/: Codex, Claude, Gemini adapters
• templates/report.md: Required final report format
• examples/: Example PR assessment input/output
• scripts/validate.ps1: Package validation

Quick start

cd "E:\Github Automation\pr-truth-gate-skill"
powershell -ExecutionPolicy Bypass -File ".\scripts\validate.ps1"

Adapter entry points:

• adapters/codex/AGENTS.md
• adapters/claude/.claude/commands/pr-truth-gate.md
• adapters/gemini/GEMINI.md

Review principles

• No evidence -> no claim
• Security and trust-boundary findings outrank style issues
• Prefer small, reversible fixes
• Block when exploitability is credible and tests are missing

License

MIT