Text-search command reference (CLI)

Course resources · class 2 exercise · class 3 exercise

Use these commands with story.txt, the course copy of 「蜘蛛の糸」. Open a terminal. The last section, Read an agent’s commands, explains the commands in the class 5 recordings.

Folders and input

Command Meaning
pwd Show the current folder’s path
ls List its contents
mkdir -p dh-b-2026 Create a working folder if needed
cd dh-b-2026 Move into that folder from its parent; run once

The download overwrites any existing story.txt in the current folder. Keep annotations in a separate file.

curl -fL -o story.txt https://soranoha.org/works/000092_000879/plaintext

-f fails on HTTP errors; -L follows redirects; -o sets the output file path.

Search and counting units

Command Output
grep --color=auto -nF 'のぼ' story.txt Matching lines with line numbers
grep -cF 'のぼ' story.txt Number of matching lines
grep -oF 'のぼ' story.txt Each occurrence on a separate line

--color=auto highlights matches in a terminal on macOS and Linux without altering matches or counts.

To count occurrences, pipe the output:

grep -oF 'のぼ' story.txt | wc -l

-F matches exact strings. A pipe (|) sends output into the next command; wc -l prints the incoming line count. Use straight quotes ('), not curved quotes (‘ ’). Output after each stage, illustrated.

The bundled text has 5 matching lines and 12 occurrences, including あのぼんやり. Counting units illustrated.

How can I save search output to a file?

Paste the command and output into your notes, or use > to save the output:

grep -nF '落ち' story.txt > down.txt

> writes the output to down.txt, replacing a file of the same name. Keep the command in your notes as well.

Search with a pattern

With -E, grep reads its argument as a regular expression. It cannot be combined with -F.

Command Output
grep -onE 'のぼ.' story.txt のぼ and the next character, one per line with line numbers
grep -oE 'のぼ.' story.txt | sort | uniq -c A count for each following character
grep -oE 'のぼ[っりるれ]' story.txt | wc -l Occurrences followed by one of the bracketed characters
Symbol Meaning
. Any one character
[っりるれ] One character from inside the brackets

The bundled text has 12 matches for のぼ. (including one のぼん) and 11 for のぼ[っりるれ].

Recover from a problem

Problem Action
Download fails Save the local text copy as story.txt in the working folder
No such file Check the current folder and filename with pwd and ls
Terminal waits for more input Press Ctrl+C; check that both quotes are present
No matches Check the spelling in the source
. does not match one character and the output is garbled Check that locale shows UTF-8, and set the terminal’s character encoding to UTF-8
grep is not recognized on Windows Close Command Prompt or PowerShell; open Ubuntu on WSL
A command is unavailable Follow Set up ling; for the class 2 homework, browser search also works

Read an agent’s commands

In the class 5 recordings (Claude Code’s and Codex’s response A and recording B), the agents join several commands into one block. Read a block one part at a time, from left to right, and look up each part below. The examples are lines from those recordings; you do not need to run them.

Look at files and help

Command Meaning Example from a recording
cd Move to another folder; files are not changed cd class4-akutagawa
ls -la List every file, including hidden ones, with size and date ls -la class4-akutagawa
find . -maxdepth 3 List the files in the current folder and its subfolders, down to three levels find . -maxdepth 3
cat Print a whole file cat class4.lx
head -5, head -c 3000 Print only the first 5 lines (10 by default), or the first 3,000 bytes head -5 works.csv
sed -n 1,60p Print only lines 1–60 ling matches –help | sed -n 1,60p
cut -c1-600 Keep characters 1–600 of each line … –tsv –context 0 | head -3 | cut -c1-600
wc -l Count lines wc -l works.csv
--help Print how to use a command ling summarize --help
which Print where a program is installed; nothing if it is missing which jq
echo Print the text that follows; agents use it to label outputs echo ----

Select and reshape output

Command Meaning Example from a recording
grep -E Keep the lines that match a regular expression; inside the quotes, | means “or” grep -E 'VERB|落ちる'
cut -d, -f1-4 Keep columns 1–4 of comma-separated lines; without -d,, cut splits at tabs cut -d, -f1-4
awk '…' Run a small program on each line. In the example, for each line whose first tab-separated column is deixis, column 4 is added to the 来る total or the 行く total, depending on column 3; END prints the totals after the last line awk -F'' '$1==“deixis”{ … } END{print …}'
jq '…' Select and reshape the JSON lines that ling … --json prints. Inside the quotes, | belongs to jq, not to the shell jq -r 'select(.type==“match”) | .data | …'
python - <<'PY' … PY Run the Python program written between the two PY lines python - <<'PY'

Join commands

Symbol Meaning Example from a recording
A | B Send the output of A into B (pipe, illustrated) ling corpus inspect akutagawa.lingcorpus 2>&1 | head -40
A; B Run A, then B, even if A failed cd /work; ls -la
A && B Run B only if A succeeded curl … && unzip -l class4-akutagawa.zip
2>&1 Send error messages to the same place as the output ling corpus inspect akutagawa.lingcorpus 2>&1
2>/dev/null Discard error messages ls ~ 2>/dev/null
$(…) Replace this part with the output of the command inside file $(which ab-aozora)
for x in a b; do … done Run the commands between do and done once for each value; $x is the current value for d in 来る 行く; do … done
"…$v…" and '…' Inside double quotes, $v is replaced by its value; inside single quotes, the text stays as written --pattern "$v [lemma=て] [lemma=$d]."

Create, change, or delete files

These commands write to the disk. Find them in every block, and compare them with what the agent states that it created, changed, or deleted. /tmp is a temporary folder outside the working folder; a file written there is still a created file.

Command What it writes Example from a recording
> FILE Saves the output as FILE, replacing a file of the same name --json > class4-akutagawa/summary.jsonl
2> FILE Saves the error messages as FILE 2>/tmp/sum.err
curl -o FILE URL Downloads URL as FILE curl -sSfL -o class4-akutagawa.zip https://…
unzip Extracts the files in a ZIP; -o replaces existing files without asking. unzip -l only lists the contents and writes nothing unzip -q class4-akutagawa.zip
mkdir -p Creates a folder mkdir -p /tmp/aoz_dl
cat > FILE <<'EOF' … EOF Saves the lines between the two EOF lines as FILE cat > /tmp/aoz_scripts/list.py <<'EOF'
rm Deletes files without asking rm /tmp/sum.tsv /tmp/sum.err

One block, part by part

In Claude Code’s recording B, this block counts matches and works for each combination:

for d in 来る 行く; do for v in '[upos=VERB]' '[lemma=落ちる]'; do echo "== $v + て + $d"; ling summarize --corpus akutagawa.lingcorpus --pattern "$v [lemma=て] [lemma=$d]." --json --by-document | jq -s '{matches: (map(select(.type=="complete"))[0].data.counters.matches_found), works: (map(select(.type=="summary")|.data.document)|unique|length)}' -c; done; done
Part What it does
for d in 来る 行く; do … done Runs the inner part twice, with $d set to 来る and then 行く
for v in '[upos=VERB]' '[lemma=落ちる]'; do … done Inside that, runs twice more, with $v set to any verb and then 落ちる: four runs in all
echo "== $v + て + $d" Prints a label such as == [lemma=落ちる] + て + 来る
ling summarize … --pattern "$v [lemma=て] [lemma=$d]." Counts one pattern, such as [lemma=落ちる] [lemma=て] [lemma=来る].; --json --by-document prints the counts as JSON lines, one for each work
| jq -s '{matches: …, works: …}' -c Reads all the JSON lines at once (-s) and prints one line (-c): matches is the total number of matches, and works the number of different works that contain one

The block writes no files. Its output, with a label line before each result:

== [upos=VERB] + て + 来る
{"matches":553,"works":66}
== [lemma=落ちる] + て + 来る
{"matches":10,"works":6}
== [upos=VERB] + て + 行く
{"matches":318,"works":56}
== [lemma=落ちる] + て + 行く
{"matches":1,"works":1}

The unit of matches is matched token sequences, and the unit of works is works with at least one match.

Search the course pages