Week 02: CLI Fluency I — Files, Pipes and Text

Phase 1 · Ways of working: the AI-native team

Domain pack: Using a domain pack? Do the challenge on a public API from your domain as well, for example openFDA for pharma. See the domain packs.

1. Concept brief

Unix tools each do one thing and pass text to each other. Fluency means chaining small tools instead of opening a GUI. This week you learn to move around, inspect files, and transform text.

Vocabulary

stdin / stdout / stderr: the three default streams. > redirects stdout to a file, 2> redirects stderr, | pipes stdout into the next command’s stdin.
Absolute vs relative path: /Users/you/x vs ./x. ~ means your home folder; .. means the parent folder.
Permissions: rwx for user, group and others. chmod +x makes a file executable.
Glob: a shell wildcard (*.csv) that the shell expands before the command runs.
Exit code: 0 means success; anything else is failure. $? holds the last exit code.
JSON: the default data format of APIs. jq is how you query it.

2. Guided lab (~3 hrs)

mkdir -p ~/dev/product-engineer-lab/week-02 && cd $_

# Navigation and inspection
pwd; ls -la; tree -L 2 ~/dev
cat /etc/hosts; less /etc/services   # press q to quit
head -n 5 /etc/services; tail -n 5 /etc/services
wc -l /etc/services

# Redirection and pipes
echo "hello" > a.txt; echo "world" >> a.txt; cat a.txt
ls /nope 2> errors.txt; cat errors.txt
cat /etc/services | grep -i http | wc -l

# Search
rg "tcp" /etc/services | head
find ~/dev -name "*.md" -type f
find . -name "*.txt" | xargs wc -l

curl -sL https://raw.githubusercontent.com/datasets/airport-codes/main/data/airport-codes.csv -o airports.csv
head -3 airports.csv
cut -d, -f2,3 airports.csv | sort | uniq -c | sort -rn | head

# JSON with jq
curl -s https://api.github.com/repos/astral-sh/uv/releases?per_page=5 > releases.json
jq '.[0] | keys' releases.json
jq -r '.[] | "\(.tag_name)\t\(.published_at)"' releases.json
jq '[.[] | {tag: .tag_name, assets: (.assets | length)}]' releases.json

# Permissions
echo 'echo "I ran"' > run.sh; ./run.sh   # fails — why?
chmod +x run.sh; ./run.sh; ls -l run.sh

3. Unguided challenge (~3 hrs)

Generate a fake support-ticket log with this script (save it as gen.sh and run bash gen.sh > tickets.log):

for i in $(seq 1 2000); do
  s=$(shuf -n1 -e open pending resolved escalated closed)
  echo "$(date -v-${RANDOM:0:2}M +%FT%T) ticket_id=$i account=A$((RANDOM%50)) status=$s response_s=$((RANDOM%600))"
done

(On macOS, run brew install coreutils first and use gshuf if shuf isn’t found.)

Using only CLI tools, answer the questions below. Save each command in answers.sh with a comment:

How many tickets per status?
The top 5 accounts by escalated count.
The average response_s of resolved tickets (hint: awk).
Every ticket with response_s over 500, written to slow_tickets.log.
A JSON array of {account, escalations} objects, built with jq -R / jq -s.

Acceptance criteria: each answer is a single pipeline, and bash answers.sh reproduces all the results.

4. Ways of working (~1 hr)

The AI-native team. Small pods (2–4 people), builder-PMs who prototype, and agents doing first drafts of code, research and tests. Read two current write-ups on how AI-native teams are organised; search for recent posts from Linear, Vercel or Anthropic on how they build. Then write: what does a PRD become when a prototype takes an afternoon? What gets more important, and what disappears?

5. Check

What’s the difference between > and >>?
Why does ls *.csv behave differently in an empty folder?
What does 2>&1 do?
Write a jq filter that returns the name of every object in an array.
What does exit code 127 usually mean?

Recycled from week 1:

What is PATH?
What does idempotent mean?

6. Teach-back

Write “Pipes explained with a factory line metaphor” (200 words), with three real one-liners from this week.

Done when

Guided lab completed
answers.sh reproduces all five answers
Personal CLI cheat sheet (cheatsheet.md) committed
Ways-of-working note written
Check and teach-back done