Add handoff skill for session continuity

[Mrbaeksang]

What

Adds a new cross-cutting handoff skill that compacts the current conversation into a resume-ready document, so another agent or a future session can continue the work without re-deriving context.

This fills a genuine gap in the pack: there is currently no skill for preserving work state across sessions / compact / agent handoffs. It complements context-engineering (which loads context in) by handling the out direction.

Why a skill, not prose

It follows the standard anatomy from docs/skill-anatomy.md:

• Overview / When to Use — including explicit "When NOT to use" (task complete → write a commit/PR instead; everything already in durable artifacts → reference them)
• Process — choose an OS temp-dir destination (not the workspace), capture only conversation-local state, reference durable artifacts instead of duplicating them, redact secrets/PII, tailor to the next session's focus
• Common Rationalizations / Red Flags — anti-patterns like transcribing instead of compacting, committing the doc into the repo, omitting rejected approaches
• Verification — evidence checklist

Adapted from Matt Pocock's handoff skill, reworked into a model-agnostic, process-driven form (attribution noted in the skill).

Integration

• skills/handoff/SKILL.md (new)
• using-agent-skills: discovery flowchart branch, lifecycle note, quick-reference row
• README.md: skill count 24 → 25, skill table row, project tree
• CLAUDE.md: cross-cutting phase entry
• AGENTS.md: intent → skill mapping

Verification

Per CONTRIBUTING.md, this PR touches skills/using-agent-skills/SKILL.md (embedded by the session-start hook), so the hook regression test was run:

$ bash hooks/session-start-test.sh
session-start JSON payload OK

Frontmatter validated: name matches the directory, description is what-then-"Use when" and is 320 chars (< 1024).

🤖 Generated with Claude Code