@hivehub/rulebook 4.4.1 → 5.0.0

package/templates/rules/task-decomposition.md

@@ -0,0 +1,32 @@
+---
+name: task-decomposition
+tier: 2
+description: "Decompose multi-file tasks into 1-2 file sub-tasks"
+alwaysApply: true
+filePatterns: ["*"]
+tools: ["all"]
+---
+
+# Task Decomposition
+
+When a task touches multiple subsystems, decompose into sub-tasks where each modifies **at most 1-2 files**.
+
+## Why
+
+AI agents lose accuracy when editing 3+ files in one pass. Context from earlier files gets compressed away.
+
+## Rules
+
+1. **Each sub-task modifies 1-2 files** — never more
+2. **Each sub-task is independently verifiable** — it compiles, it doesn't break existing behavior
+3. **Sub-tasks follow data flow order** — upstream first (component → buffer → renderer → shader)
+4. **Each sub-task has a clear "done when"** — not "implement X", but "field Y exists with default Z"
+5. **Build after each sub-task** — verify compilation before proceeding
+
+## When an Agent Receives a Multi-File Task
+
+1. **STOP** — do not start implementing
+2. **Create a change plan** listing all files and dependency order
+3. **Decompose** into sub-tasks following this rule
+4. **Report back** with the decomposition
+5. **Implement** one sub-task at a time