A disciplined CLI workflow for AI-assisted software development: PRD → Design → Plan → Implement → QA → Compliance → Archive. Human-driven by design (L1 self-application) — every transition is operator-approved, every artifact is auditable. Datarim is the operational backbone of the Arcanada ecosystem; the same lifecycle that ships Verdicus also ships Datarim itself.
Capabilities
- Full lifecycle: /dr-prd, /dr-design, /dr-plan, /dr-do, /dr-qa, /dr-compliance, /dr-archive
- Thin-index task schema (tasks.md + per-task description files)
- Reflection + evolution loop with Class A/B proposal gating
- Stack-agnostic gate enforcement — no language or framework lock-in
- Self-healing operational files via /dr-doctor (schema migration, orphan detection)
- 20+ skills on disk (factcheck, humanize, ai-quality, research-workflow, …)
- Multi-agent workspace discipline (selective git add -p, blob-swap recipe)
- Quarterly /dr-optimize sweep — bloat detection and stale-validation flags
Current autonomy level
L1
Weakest link
Human-driven by design — L1 is the target, not a gap. The only operational risk is index drift between tasks.md and on-disk task description files; mitigated by /dr-doctor but not yet auto-detected post-commit.
Roadmap to L2
- L2 lift — minimal /health for the framework runner; per-task post-completion event POSTed to Ops Bot.
- L2 polish — git pre-commit hook running /dr-doctor in --check mode to catch index drift before push.
- Verification gate — 30 days without manual /dr-doctor remediation; framework version pin enforced across all projects.