The tool catalogue
This list is generated from the app's own registration code, so it is what PounceCode actually registers rather than a description of it. Tools marked asks before running are treated as dangerous: they go through a confirmation before they touch anything, and they are refused outright while a plan is waiting for your approval.
How loading works
A model choosing from ninety tools chooses badly — and the smaller the model, the worse it gets. So PounceCode registers a bootstrap set that is always present, declares the rest as named groups, and gives the model two tools for browsing and loading them.
What is always there
The task board, file reads and writes, edits, shell, git and the host probe. Git is in this set deliberately: nearly every coding request touches it, and spending a discovery round to ask for git_status is a round not spent on the work.
What has to be asked for
Web, batch, processes, images, CSV — and any MCP servers you have configured. The prompt states plainly that the visible set is partial, and an unknown-tool error names the group that holds the tool, so a wrong guess costs one round rather than a dead end.
Limits you can set
Every read tool has a ceiling, so one read_folder on a big repository cannot swallow the context window. The defaults are sensible; all of them are configurable per profile.
| Limit | Applies to |
|---|---|
max_file_lines | A single file read |
max_grep_results | Search hits returned |
max_dir_entries | Directory listings |
max_shell_lines | Output kept from a command |
max_folder_files · max_folder_chars | Multi-file folder reads |
max_sql_rows | Rows returned from a query |
max_web_chars | Text kept from a fetched page |
max_image_kb · max_thumb_width | Images read for inspection |