The filesystem is the set of files an agent can read and write, and it is the most important part of its environment. For a coding agent, files are both where it learns what is true and where it makes changes that last. Almost everything an agent does meaningfully touches the filesystem at some point.
Source of truth and workbench at once
The filesystem plays two roles. First, it is the agent's best source of ground truth. Reading the actual code, the real config, and the current types beats anything the model can recall from training, because the files are reality and its memory is a guess. Pulling a file into context with a read tool is how an agent replaces assumption with fact.
Second, it is where durable work happens. A model can produce text all day, but until something is written to disk nothing has really changed. Editing a file is the agent making a change the rest of your toolchain (the compiler, the tests, git) can actually see.
Why it sits at the centre:
- Reads ground the agent in what your project really contains.
- Writes are the actual deliverable, the edits you keep.
- It persists across turns and sessions in a way the conversation does not.
A file read is not free
Every file the agent reads becomes tokens in the window. Dumping a whole directory in "just in case" crowds out room and attention. The skill is reading the right files, not all of them, and letting the agent go back for more when it needs them.
Related terms
Environment
The environment is the surroundings an agent acts in: working directory, files, shell, environment variables, and network. It defines what the agent's tools can actually reach.
Read definition →Tool
A tool is a named action, with a typed input schema, that a model is allowed to call. Tools are how a model that can only produce text gets to actually do things: read a file, run a command, search the web.
Read definition →Context
Context is all the text a model can see for a single request: the system prompt, your message, the conversation so far, and any files or tool output the agent has pulled in. It is the only thing the model knows about your specific situation.
Read definition →