Roadmap

Future milestones for the AI Guardrails project. Phase 1 (static site) is complete.

Phase 2: Autonomous update pipeline

Keep the playbook current as new AI agent tools, rules formats, and research emerge — without manual monitoring.

How it would work

Daily cron (GitHub Action)
  → Search RSS / HN / GitHub releases for new agent tools
  → Feed diffs to an LLM analyzer
  → If significance score > threshold:
      → Open draft branch + PR with suggested content changes
  → Maintainer reviews → merges → Netlify redeploys

Components

Component	Purpose
Source watchers	Monitor GitHub releases (Cursor, Claude Code, Codex, Copilot), HN AI threads, arxiv papers on AI code quality
LLM analyzer	Score each finding: "Does this change our recommendations?" (0-10)
PR generator	Draft markdown updates to guides, playbook sections, or new guide pages
Human review gate	No auto-merge — maintainer approves every change

Significance thresholds

Score	Action
0-3	Log only, no PR
4-6	Open draft PR with "consider" label
7-10	Open draft PR with "recommended" label + notify maintainer

Phase 3: Interactive setup wizard

Web-based tool that generates guardrail files for your project:

1. Pick your agent(s)
2. Pick your stack
3. List forbidden scope
4. Download a zip with AGENTS.md, rules, pre-commit config, CI workflow

Phase 4: Rule pack library

Pre-built rule packs for Python, TypeScript, Go, Rust, and mobile stacks.

Phase 5: Community contributions

Submit integration guides, share Project Learnings, vote on playbook changes.

Phase 6: Analytics (privacy-first)

Optional, opt-in pattern counts from CodeDiet — no code sent. Aggregate data improves default rules.

---

Have an idea? Open an issue on GitHub.