Clean a Prompt in a Notebook Cell with %%scikitplot#
The cell form %%scikitplot COMMAND gives the cell’s text to the command
on standard input. With cleanprompt that turns a notebook cell into a
guarded text box: what you paste is redacted before it goes anywhere, and
the vault that can restore it is written where you say.
The steps:
redact a cell, keep the vault in a folder of your choice;
restore a model’s reply with the same vault;
read what ran (exit code, output, where it ran);
the edge cases: an empty cell, a command that needs no input.
cleanprompt is a delegated command: it runs in a subprocess of the
kernel’s own Python, streamed, and stopped by the interrupt button or by the
timeout setting. Every value below is reserved for documentation.
# Authors: The scikit-plots developers
# SPDX-License-Identifier: BSD-3-Clause
A vault for this example#
cleanprompt writes the removed values to a vault. By default that is a
file in your state directory; here a temporary folder keeps the example
self-contained and leaves nothing behind.
import io
import shutil
import tempfile
from pathlib import Path
from scikitplot._ipymagic import MagicSettings, route, run_command
# In a notebook (this page's download), load the magics so the ``%%`` cells
# below run; a plain script has no IPython and skips this.
_shell = globals().get("get_ipython", lambda: None)()
if _shell is not None:
_shell.run_line_magic("load_ext", "scikitplot")
work = Path(tempfile.mkdtemp(prefix="skplt-ipymagic-"))
vault = work / "vault.json"
print("cleanprompt runs in a subprocess:", route(["cleanprompt"]).delegated)
cleanprompt runs in a subprocess: True
Redact a cell#
%%scikitplot cleanprompt redact --vault {vault}
Mail me at jane.doe@example.com or call +1 202 555 0143.
{vault} is a Python variable expanded by IPython. The redacted text is
printed; the report and the vault path go to standard error.
cell = "Mail me at jane.doe@example.com or call +1 202 555 0143.\n"
redacted = run_command(["cleanprompt", "redact", "--vault", str(vault)], stdin=cell)
print(redacted.summary())
Mail me at [EMAIL-1] or call [PHONE-1].
warning: redacted 2 value(s): EMAIL=1, PHONE=1
Placeholders are stable; restore the reply with the same vault. Note that names, organisations, places and other named entities are still not being detected.
-> spaCy is installed with its model; enable it for this run (--ner on the command line, or add spacy_detector() to the registry)
placeholder kind count detector
----------- ----- ----- -----------
[EMAIL-1] EMAIL 1 regex:EMAIL
[PHONE-1] PHONE 1 regex:PHONE
vault: /tmp/skplt-ipymagic-gpclnnbf/vault.json (overwrite, 2 value(s))
scikitplot cleanprompt: exit 0 in 0.68 s (subprocess)
The result keeps both streams, so a notebook can test the outcome instead of reading it:
assert redacted.ok and "jane.doe@example.com" not in redacted.stdout
assert vault.is_file()
print("safe to send:", redacted.stdout.strip())
safe to send: Mail me at [EMAIL-1] or call [PHONE-1].
Restore a reply#
Send the redacted text to a model; its reply keeps the placeholders. The same vault puts the values back:
%%scikitplot cleanprompt restore --vault {vault}
I will write to [EMAIL-1] and call [PHONE-1] tomorrow.
reply = "I will write to [EMAIL-1] and call [PHONE-1] tomorrow.\n"
restored = run_command(["cleanprompt", "restore", "--vault", str(vault)], stdin=reply)
expected = "I will write to jane.doe@example.com and call +1 202 555 0143 tomorrow."
assert restored.ok and restored.stdout.strip() == expected
restored 2 placeholder(s)
I will write to jane.doe@example.com and call +1 202 555 0143 tomorrow.
A list of words, not a string, is the safe way to pass a path that may hold
spaces or quotes; a string is split like a shell would ("a b" is one
word).
Edge cases#
An empty or whitespace-only cell gives the command no input at all, so
%%scikitplot cleanprompt --help behaves like the line magic:
blank_cell = " \n"
stdin = blank_cell if blank_cell.strip() else None # what the magic does
help_result = run_command(["cleanprompt", "--help"], stdin=stdin, out=io.StringIO())
print(help_result.summary())
print(help_result.stdout.splitlines()[0])
scikitplot cleanprompt: exit 0 in 0.37 s (subprocess)
Usage: python -m scikitplot.cleanprompt [OPTIONS] [COMMAND] [ARGS]...
A command that fails shows its own message and returns its exit status;
restore --strict refuses a placeholder the vault does not hold:
err = io.StringIO()
strict = run_command(
["cleanprompt", "restore", "--strict", "--vault", str(vault)],
stdin="Who is [PERSON-9]?\n",
err=err,
)
print(strict.summary())
print(err.getvalue().strip().splitlines()[-1])
scikitplot cleanprompt: exit 1 in 0.44 s (subprocess)
error: placeholder '[PERSON-9]' is not present in the vault
For a long-running command, timeout stops it (exit status 124, as GNU
timeout) together with every process it started:
%scikitplot_config timeout=120
guarded = MagicSettings(timeout=120)
print(guarded)
MagicSettings(raise_on_error=False, timeout=120, isolate=False, stream=True)
Clean up the example’s folder (the vault holds the removed values):
shutil.rmtree(work)
Total running time of the script: (0 minutes 1.916 seconds)
Related examples
Cleanprompt Basics: Hide a Value, Then Get It Back
Inspect and Self-test the MCP Server from a Notebook
