A context pointer is a reference you hand the agent in place of the actual content: a file path, a URL, a ticket id, a function name. Rather than pasting a 2,000-line file into the prompt, you point at it and let the agent open it if the task requires.
Cheap context
The appeal is cost. A pointer is a few tokens; the thing it points to might be thousands. By passing the reference instead of the payload, you keep the context window lean and let the agent pull in only what it actually uses. This is the mechanism behind progressive disclosure: breadth on tap, not breadth resident.
Pointers work because a coding agent can act. Give it a path and it reads the filesystem; give it a URL and it fetches the page; give it a symbol and it greps the repo. Each pointer resolves to a primary source at the moment of need.
- A file path instead of the file's contents.
- A URL instead of a pasted article.
- An issue id instead of the whole thread.
Related terms
Progressive disclosure
Progressive disclosure is revealing detail to the model only when it is needed, via pointers and on-demand loading, instead of putting everything into context up front. It saves window space and attention.
Read definition →Primary source
A primary source is the authoritative original: the actual code, the real types, the official docs. Point agents at primary sources so they read reality instead of guessing from memory.
Read definition →Filesystem
The filesystem is the set of files an agent can read and write. It is its main source of truth and its main way to make durable changes.
Read definition →