Dots + Opus 5.5: how to make your Dots team better (Full Guide)
Short summary
A practical guide to connecting Claude Opus 5.5 as an external tool to ChatGPT's Dots system, using it selectively for deep research, strict reviews, visual analysis, and long-running tasks while keeping Dots as the orchestrator. The article walks through 10 specific upgrades to make the combined system cheaper and more capable.
- Opus 5.5 connects as a private backend tool called only when tasks need deeper context, stricter review, visual analysis, or patience — not as a model replacement inside Dots itself.
- The four weakest points in Dots (shallow reading, uniform effort levels, weak self-review, poor visual work) map directly to where Opus adds value, and each has a specific upgrade to fix it.
- Caching the unchanging prompt prefix (standing brief, house rules, schemas) with a one-hour TTL cuts costs dramatically for teams running multiple tasks per day, and compacting the context window before it fills prevents mid-task cutoffs.
- Sorting work by whether anyone is waiting lets you move patient tasks (weekly digests, backlog triage, doc sweeps) to off-hours at lower cost, and moving house rules into the repo eliminates contradictions across four separate places.
Full summary
The core idea is using Claude Opus 5.5 as a specialized backend tool for your ChatGPT Dots workflow, not replacing Dots but filling its weak spots. Dots handles orchestration reliably but struggles with deep analysis of complex sources, visual content, and nuanced review work. The setup involves building a private ChatGPT app with one tool called `run_opus_task` that routes requests to an Opus backend you control, keeping your Anthropic API key secret and starting read-only. Your dot stays in charge—it decides when to call Opus for deeper context, stricter reviews, or visual analysis, then continues the conversation with the results.
The ten upgrades are practical fixes to make this pairing work better. Scout adds a scheduled deep read on top of daily skims, targeting the "what's quietly going wrong" questions that need full context. Builder gives the model a gear lever so tasks get appropriate effort levels instead of everything running at the same speed. Reviewer separates the review from the work itself and uses strict JSON schemas that can't pad answers with excuses. Courier makes visual reading mandatory before filing—dashboards, screenshots, and diagrams get actual analysis instead of vague descriptions. These four upgrades directly address where Dots gets thin.
The remaining upgrades handle the operational side of running agentic loops at scale. Stop resending identical context by ordering prompts so unchanging material comes first, then use a one-hour TTL cache so multiple runs share the same prefix. Compact messages proactively before the context window fills, not after—put the block first in messages and remove what it summarized. Give the loop a visible budget so it paces itself instead of getting cut off mid-task. These aren't flashy features, but they're what keeps a workflow from degrading as it runs longer and hits real constraints.