Docs › Agent
What the agent can do
The tools the agent has: reading, editing, searching, running commands, browsing, and calling your connectors.
In Agent mode the model works by calling tools. Each call appears in the transcript, and read-only calls are folded into a single Explored line you can expand.
The tools
| Tool | What it does |
|---|---|
| read_file | Reads a file from the workspace. |
| list_files | Lists files matching a glob, across every open folder. |
| search_files | Searches file contents with ripgrep. |
| edit_file | Changes part of a file. Shows a diff before applying. |
| write_file | Writes a whole file. Shows a diff before applying. |
| run_terminal_command | Runs a shell command. See [Terminal](/docs/agent/tools/terminal/). |
| Browser tools | Navigates a page, reads the DOM and console, clicks. For checking your own running app. |
| web_fetch | Fetches a URL and returns its text. |
| Connector tools | Anything your MCP servers expose. See [Connectors](/docs/mcp/overview/). |
It checks its own work
Before an agent run says it is finished, AstraCode runs your project's own check (npm test, cargo test, go test ./..., pytest or make test), and if it fails, hands the output back to the model with one attempt to fix it. If it still fails, the run ends and says so.
The command is detected, never guessed: it only runs one your project actually declares. It runs once per run, only in Agent mode, and only when files were changed. Turn it off with astracode.agent.verify, or override the command with astracode.agent.verifyCommand if your full suite is too slow.
This exists because an agent marking its own todo list complete is not evidence of anything. A failing test is.
Todos and plans
For anything with more than a couple of steps the agent writes a todo list and ticks items as it goes. If it stops with steps still open, AstraCode tells it so, twice, before letting the run end.
In Plan mode the list becomes a plan document saved in your repository (or your home folder, if you prefer, astracode.plan.location).
All documentation
Get started
Agent
- Agent overview
- Approvals and permissions
- Plan mode
- What the agent can do
- Running commands
- Browser tools
- Checkpoints and undo
- Infrastructure as code
- Running several agents at once