task(skills): remove curated subagent-driven-development, preserve context-management guidance

[pentaxis93]

Parent: #44

Summary

Investigation of the curated subagent-driven-development skill from obra/superpowers found no real exigence for a standalone skill. Every pattern it claims is either default tool behavior (fresh subagents already start clean), already covered by existing skills (verification-before-completion, test-first, CI review), or basic prompt engineering advice. The skill has zero pipeline integration — no cross-references from any native skill, and all its own cross-references point to uninstalled superpowers skills.

Direction: removal, not replacement

Two genuinely useful insights survive and are preserved as integrated guidance (not a standalone skill):

1. Use subagents to manage context across multi-task execution — dispatching fresh subagents prevents context pollution as work progresses through a plan.
2. Choose the subagent model thoughtfully — match model capability to task complexity. Use cheaper/faster models for straightforward tasks; reserve the most capable model for subtle or complex work.

These are added to using-groundwork stage 4 and WORKFLOW.md stage 4.

Acceptance criteria

• subagent_driven_development entry removed from agents.toml
• Skill block removed from skills/skills.toml
• Installed copies removed from .claude/skills/ and .codex/skills/
• subagent_driven_development removed from CLI skill list in main.rs
• using-groundwork stage 4 updated with context-management and model-selection guidance
• WORKFLOW.md stage 4 updated: SDD paragraph replaced with context-management note, routing table entry removed
• README.md skills table: SDD row removed
• CHANGELOG entry added
• grep verification: zero matches outside CHANGELOG history and docs/research/

Dependencies

Depends on: #44 (evaluation verdict drives this work)

[pentaxis93]

Implemented and merged in PR #110 (commit 89f6822). Direction changed from replacement to removal after investigation found no exigence for a standalone skill. Two useful insights (fresh-subagent context management, model-selection) preserved as integrated guidance in using-groundwork and WORKFLOW.md stage 4. Closing as complete.