Context Handoff Protocol

Every AI tool runs its own context window. The moment you copy raw output from Perplexity into Claude — or from one Claude session into the next — you're making a lossy transfer. The rejected options, the source confidence levels, the questions you explored and dismissed, the follow-ups that shaped your thinking — all of it falls away. The receiving tool sees a flat text block with no history and produces a flat, generic response. People then blame the receiving tool, when the real failure was at the handoff.

This skill is a set of structured templates that take roughly 60 seconds to fill and save 10 minutes of re-explaining (or worse, accepting a weaker downstream output). There's a generic handoff that works for any tool pair, plus specific templates for the common ones: Perplexity → Claude (research to synthesis), Claude → Claude Code (planning to implementation), Claude Code → Claude (implementation to review), ChatGPT → Claude (exploration to execution), and same-tool session continuation. Each template enforces the five rules of clean handoffs — transfer decisions not just outputs, flag uncertainty explicitly, specify what the receiving tool should do, keep handoffs under 500 words, and use files for persistence rather than chat for anything you'll need next week.

The Claude → Claude Code template is the one that quietly changes the most. It captures rejected architectural approaches as well as chosen ones — because without them, Claude Code may suggest the exact thing you already ruled out. The Perplexity → Claude template captures source-credibility judgements and known gaps, so synthesis weights strong findings appropriately and flags real conflicts instead of laundering them into a smooth narrative.

It also names the cases where you should not use a structured handoff. Simple factual transfers (a single URL, a single number) don't need one. Quick one-off tasks under five minutes don't need one. And sometimes you want context loss deliberately — for editing, code review, or test-case generation — where a fresh perspective is the entire point. The skill calls that out explicitly so you don't over-engineer transfers that should stay simple.