Agent harness
The app around the model — VS Code, CLI, cloud agent — that hands it files, tools and UI.
One page for co-developers working with coding agents.
The app around the model — VS Code, CLI, cloud agent — that hands it files, tools and UI.
Ask mode explains without touching code · Edit mode changes one file fast · Agent mode plans, edits across files, runs commands and self-corrects.
The model's working memory: every instruction, file and tool result it can see this turn.
The provider's baseline instructions — invisible to you, sent with every single request.
Persistent project rules the agent re-reads every session, so you don't repeat yourself.
Packaged expertise — instructions plus assets. Only the short description sits in context until the task needs it.
The files you attach yourself — the fastest way to point the agent at the right code.
Bundles that extend the harness — skills, tools and workflows shipped in one install.
The concrete actions it can take: read, edit, run a command, search, fetch a URL.
A standard way to connect the agent to other systems — package feeds, ticket trackers, databases.
The red overlap is the exploitable case: reads your secrets + reads an untrusted ticket + can act outward → one poisoned issue is enough. Break the overlap and the attack dies.
Untrusted content the agent reads — web pages, issues, third-party READMEs — can smuggle instructions into the context and be mistaken for your orders.