@parke.dev/pi-file-search

First-class fd and rg tools for the pi coding agent. Faster and more predictable than asking the model to compose find / grep shell pipelines: both respect .gitignore by default, output is bounded and spilled to a file when it overflows, and user-supplied patterns are never interpreted as command-line flags.

Two tools:

Tool	What it does
fd	find files and directories by name (regex by default; optional glob)
rg	search file contents with ripgrep (file:line:match)

Ships with a file-search skill that teaches the model to prefer these over bash find / bash grep.

Install

pi install npm:@parke.dev/pi-file-search

You also need fd and rg on the machine (or dropped into the agent bin dir — see below). On macOS and Debian/Ubuntu:

# macOS
brew install fd ripgrep

# Debian / Ubuntu (fd ships as fdfind; the tool probes both names)
apt install fd-find ripgrep

Restart pi after installing so the extension re-probes.

Binary provisioning

Each tool is resolved once per process, in this order:

1. System PATH — for fd, candidates are fd then fdfind (Debian's package name); for rg, just rg. Each is probed with --version.
2. Agent bin dir — ~/.pi/agent/bin/fd or ~/.pi/agent/bin/rg. The file must exist and respond to --version.

There is no automatic download. If neither location has a working binary, the tool fails with a one-line install hint (brew install … / apt install …, or “drop a binary at ~/.pi/agent/bin/<tool>”) and asks you to restart pi.

Any platform where you can put a working fd/rg on PATH or in ~/.pi/agent/bin is supported. There is no per-OS binary map and no cache of downloaded releases — install the binaries yourself (package manager, or a static build from the upstream GitHub releases into ~/.pi/agent/bin).

Tools

fd — Find Files

Find files and directories by name. Respects .gitignore and skips hidden files by default. Pattern is a regex unless glob: true.

Parameter	Type	Default	Meaning
pattern	string	list everything	name pattern (regex unless glob)
path	string	session cwd	directory to search (~ and a leading @ are expanded)
type	"file" | "directory" | "symlink"	any	restrict to one entry type
extension	string	—	filter by extension, e.g. ts (leading dot optional)
glob	boolean	false	treat pattern as a glob instead of a regex
hidden	boolean	false	include hidden and ignored files (--hidden --no-ignore)
max_depth	integer 1–64	64	max directory depth
limit	integer 1–10000	1000	max results

rg — Search Content

Search file contents. Respects .gitignore, skips binaries, returns file:line:match. Pattern is a regex unless fixed_strings: true.

Parameter	Type	Default	Meaning
pattern	string	required	regex to search for (or literal if fixed_strings)
path	string	session cwd	file or directory to search (~ and a leading @ are expanded)
glob	string	—	only search paths matching this glob, e.g. *.ts
file_type	string	—	ripgrep type, e.g. ts, py, rust
case_sensitive	boolean	smart-case	true = case sensitive, false = ignore case; omit for smart-case
fixed_strings	boolean	false	treat pattern as a literal string
hidden	boolean	false	include hidden and ignored files (--hidden --no-ignore)
context	integer 0–20	—	lines of context around each match
limit	integer 1–1000	100	max matches per file

Large result sets are truncated for the model and the full output is written to a temp file whose path is included in the response.

License

MIT