alpacalyzer-algo-trader: Skill for Claude Code

name: issue-decomposition description: Decompose features into well-formed, agent-sized GitHub issues with dependency ordering and orchestrator metadata

---

Issue Decomposition Skill

How to break features into issues that the orchestrator can dispatch to agents effectively.

Scope

This skill produces GitHub issues, not code. Output goes to gh issue create commands.

The Right Granularity

An agent-sized issue has these properties:

• One session: An agent can complete it in a single worktree session (typically 1-5 grind loop iterations)
• Focused scope: Touches 1-4 files across 1-2 modules. If it touches more, split it.
• Testable: You can write a concrete test that verifies the issue is done
• Independent merge: The PR can be merged without waiting for unrelated work
• Clear boundaries: Another agent working on a sibling issue won't have merge conflicts

Size Guide

Complexity	Agent time	Typical scope	Examples
small	< 30 min	1-2 files, single module	Add a config field, write a utility function, fix a bug
medium	30 min - 2 hours	2-4 files, 1-2 modules	New strategy implementation, new API endpoint, new scanner
large	> 2 hours	5+ files, 3+ modules	Must be decomposed further

Dependency Patterns

Common Dependency Chains

Data model → Service that uses it → CLI/API that exposes it → Integration test
Config schema → Feature that reads config → Tests
Shared utility → Multiple consumers (parallel after utility lands)

When to Use Dependencies

• Use Depends on when Issue B literally cannot start until Issue A's PR is merged (e.g., B imports a class that A creates)
• Don't use Depends on for soft ordering preferences. If B could technically start before A merges (maybe with a stub), they're parallel-safe.
• Minimize dependency depth. A chain of A→B→C→D→E means 5 sequential waves. Look for ways to parallelize: can C and D both depend on B independently?