Datarim System
Core Datarim rules. Load this entry first, then only the fragment needed for paths, storage, numbering, backlog, routing, or archive behavior.
Overview
Datarim System is the foundational skill loaded by every agent at every stage. It defines where files live, how tasks are numbered, how backlog works, and how complexity determines the pipeline route. The skill is split into a short index entry and focused supporting fragments to reduce context waste while preserving the full system contract.
File Organization
All workflow state lives in datarim/ at the project root (added to .gitignore). Completed task archives go to documentation/archive/ (committed to git). This separation lets multiple developers work in parallel with their own local state while sharing project knowledge through archives.
Task Numbering
Every task gets a permanent ID in the format {PREFIX}-{NNNN} (e.g., DEV-0042, WEB-0003). The ID is assigned once and remains unchanged across its entire lifecycle: backlog, active tasks, archive, and backlog-archive. Prefix is selected from a registry of project and area prefixes.
Complexity Routing
- Level 1 (Quick Fix) — single file, <50 lines. Route: init, do, reflect, archive.
- Level 2 (Enhancement) — 2-5 files, <200 lines. Route: init, plan, do, reflect, archive.
- Level 3 (Feature) — 5-15 files, 200-1000 lines. Route: init, PRD, plan, design, do, QA, reflect, archive.
- Level 4 (Major Feature) — 15+ files, >1000 lines. Route: init, PRD, plan, design, phased-do, QA, compliance, reflect, archive.
Fragment Routing
Agents load only the fragment they need for the current sub-problem: path-and-storage for file locations and archive boundaries, task-identity-and-context for numbering and active task tracking, model-assignment for model/effort frontmatter, backlog-and-routing for complexity levels and mode transitions, command-and-archive-rules for /dr- namespace and critical invariants.
Key Rules
- Path resolution is mandatory before any write — walk up the directory tree to find
datarim/ - All reports must include the task ID in the filename
- Backlog uses a two-file architecture: active items in
backlog.md, historical items inbacklog-archive.md - All Datarim commands use the
/dr-prefix to avoid conflicts with built-in Claude commands - Use
$HOME/.claude/or project-relative paths, not absolute machine-specific paths