Best practice for plan-to-implementation context handoff across /clear?

After plan mode produces a plan file (~/.claude/plans/*.md), what works best for carrying context into implementation — especially across /clear or new sessions?

Option A: Inject the entire plan file content into the new session

Option B: Extract key information from the plan file into structured fields and re-inject those

If Option B, should the description be copied verbatim from the plan file, or summarized/restructured?

Current approach (Option B)

I extract from the plan file into an issue tracker with these fields:

• Description: problem + solution + context + non-goals (the "what and why")
• Acceptance Criteria: Given/When/Then scenarios derived from the description (the "definition of done")
• Design: approach + edge cases + QA steps + file paths + implementation steps derived from description + AC (the "how")

Each field is derived from the previous one (description → AC → design) to maintain consistency.

Questions

1. Has anyone benchmarked which approach gives Claude better implementation accuracy?
2. Is there a recommended pattern for plan-to-implementation handoff?
3. How much implementation detail is optimal in the design field — too little and Claude improvises, too much and it follows mechanically even when code doesn't match?

The /clear case is the hardest since there's zero conversation history to fall back on.