Copy-pasteable prompts for directing AI coding agents in HubSpot.
A live HubSpot account is the hardest place to point an agent — the truth lives in the account, not your files, and there's rarely a clean undo. Each of these is a prompt we actually use — the public twin of a tool from the ops kit we run on real projects, run against HubSpot as the hard case. Copy one, answer the questions it asks, and put it to work. No signup.
HubSpot portal X-ray
twin of /system-mapHubSpot has no "find all references." This prompt makes your agent build the dependents checklist the platform won't hand you — split into what's provable from files and what only you can verify live — before you change anything.
Get the promptHubSpot conventions file
twin of /rememberGenerate the conventions file your agent reads before touching a CMS-React theme — the project shape, the code-vs-content seams, and the load-bearing don'ts that keep it from clobbering live work.
Get the promptHubSpot build brief
twin of /briefMake the agent interview you before it changes a live portal — surfacing the questions about live content, hidden dependents, and rollback that decide whether a change is actually safe.
Get the promptAn island, not a rewrite
twin of /buildDirect your agent to add interactivity to a CMS-React module as a focused island — server-rendered everywhere else, hydrated only where the feature needs it, with nothing unserializable crossing the boundary.
Get the promptA prompt is a tip. The judgement behind it is the job.
Each prompt is one move from a larger practice: directing a coding agent safely inside a live portal — mapping the blast radius before you touch anything, and proving a change landed without breaking what already worked. The course turns the moves into a habit.