When Claude Code sessions get interrupted or balloon past a few hundred messages, resuming with the full transcript burns context on stale state. This skill extracts just the actionable bits from local .claude files: the last compact summary, pending work, errors, subagent states, and files touched. It classifies how the session ended (clean exit, interrupted, error cascade, abandoned) so you know whether to retry tool calls or start fresh. The bundled Python script handles compact-boundary parsing and noise filtering automatically. Good for recovering multi-agent workflows or continuing work without replaying 5MB of resolved issues. Does what claude --resume does, but selectively.
npx skills add https://github.com/daymade/claude-code-skills --skill continue-claude-work