Multi-Tool Orchestration

Most multi-tool workflows aren't designed — they accreted. Someone added Perplexity for research, kept Claude for writing, picked up Cursor for code, and now the working pattern is whatever copy-paste glue holds it together. That glue is where quality dies. Each tool transition loses context: the reasoning behind the output, the confidence levels of findings, the gaps the previous tool noticed but didn't fill. This skill exists to design the handoffs deliberately rather than letting them happen by accident.

The process starts by mapping the workflow phases — research, analyse, plan, create, review, deliver — and assigning the tool that's genuinely best at each. Perplexity for citations. Claude for long-context reasoning and prose. Claude Code for in-codebase work. Midjourney or DALL-E for images. Native tools for the final delivery step. The match matters less than the handoff design between them.

The skill gives concrete patterns for the workflows people actually run: research-to-document, explore-plan-build for software, content pipeline with a fresh-eyes editing session, data analysis pipeline, design-to-code, and session-to-session handoff inside a single tool when you hit context limits. Each pattern specifies what artifact transfers, what format it should be in, what context to deliberately exclude, and what failure point to watch for.

It's also honest about when to use fewer tools. If you're at four or more in a single workflow, the skill challenges that count — every transition is a context loss point, and three tools orchestrated well usually beat six orchestrated badly. The anti-patterns it names are the ones nobody admits to: the round trip (A→B→A on the same task), the copy-paste marathon, tool maximalism (four tools when two would work), and context overloading (dumping all of A's output into B instead of filtering). Sometimes the skill also recommends breaking context on purpose — for editing and review, withholding the original reasoning so the reviewer evaluates the output fresh. Context preservation is the default; context breaking is a deliberate tool, not an accident.