You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A reference repository implementing stages 1 to 5 of the Trusted OSS โ AI Coding Strategy as files.
Stages 3 to 5 run as GitHub Actions workflows; stage 2 and stage 4c are config files.
Fork it for immediate use, or copy individual config files into your existing project.
Semgrep/grype findings โ AI validation & interpretation โ PR comment
PR (auto-activates when ANTHROPIC_API_KEY is set)
4b
ai-fuzzing.yml
Claude + requests
AI-generated edge cases to detect 5xx errors
Push to main ยท weekly
4c
.mcp.json, .claude/settings.json
Claude Code settings
MCP server allowlist, secret reads blocked, human approval for egress and deploy commands
Every agent session
Stage 4c governs the agent on a developer workstation rather than CI, so it ships as config files rather than a workflow.
The rules and the server intake procedure are in CLAUDE.md.
Stage 5: Continuous Monitoring & Auto-Remediation
Workflow / Config
Tool
Role
Schedule
dependabot.yml
Dependabot
Auto-generate dependency update PRs
Weekly
renovate.json
Renovate
Auto-merge critical patches
Immediately ยท weekly
dast.yml
OWASP ZAP
Dynamic vulnerability scan after deployment (soft fail)
Push to main
Quick Start
1. Fork
Fork this repository on GitHub, then clone it.
git clone https://github.com/YOUR-ORG/ai-coding-best-practice.git
cd ai-coding-best-practice
Opening a PR on GitHub will automatically trigger all 6 Stage 3 workflows.
3. Register GitHub Secrets
Secret Name
Purpose
Required
ANTHROPIC_API_KEY
Stage 4 AI review (ai-review.yml), AI fuzzing (ai-fuzzing.yml)
Optional
Once ANTHROPIC_API_KEY is registered, the Stage 4 AI defense layer activates automatically.
No additional configuration needed โ findings-driven AI review starts on the next PR.
Customization
File
What to Modify
CLAUDE.md
Team license policy, prohibited package list, MCP server intake procedure
.cursorrules
Per-tool rule adjustments
.mcp.json
Approved MCP servers (version pin required)
.claude/settings.json
Agent permissions โ blocked read paths, commands that require approval
.grype.yaml
Vulnerability threshold (high โ critical)
.gitleaks.toml
Add organization-internal pattern exceptions
.semgrep.yml
Add language/framework-specific rulesets
renovate.json
Auto-merge scope, update schedule
dast.yml
Switch to hard fail by changing fail_action: true after stabilization