Run the demos

Page 3 of 7

No AWS account. No API key. Node 18 or newer and bash, nothing else.

Step 1. Run everything

From the root of this repo:

./run-all-demos.sh

One command. Every chapter finds real work in seeded data, and the cloud chapter runs offline against canned AWS responses. Takes about a minute.

A demo that prints findings and exits non zero has not failed. That is the check saying "there is work here." non zero is the whole point. See what a loop is.

Step 2. Read one piece of output

Look for this block in what scrolled past:

Loop 5. Dead code and unused dependencies   (pack check: `knip`)
  DEAD FILE   src/unused.js  (exported but nothing imports it)
  UNUSED DEP  is-odd  (in package.json, never imported)
  The loop would remove these, then prove the build and tests still pass.

That is a real check, on a real project, finding real problems. The loop's job is the last line.

Step 3. Run one chapter on its own

Each chapter has a playground you can run by itself:

cd loop-packs/engineering-loops/demo-app && node loops.js all
cd loop-packs/ai-ml-loops/demo-app       && node loops.js all
cd loop-packs/qa-loops/demo-app          && node loops.js all

Cloud has no demo app, because it runs against AWS. Use the dry run instead:

cd loop-packs/cloud-loops
DRY_RUN=1 bash loops/15-cost-and-spend-anomaly-watch/check.sh

Step 4. Run a single check

This is the smallest unit in the whole repo. One loop, one question.

cd loop-packs/engineering-loops/demo-app
bash ../loops/02-dependency-upgrades/check.sh
echo "exit code: $?"

Exit 1 means it found work. That is the moment a real run would wake an agent.

Three loops ship working code

Most loops are orders plus a check, ready to point at your stack. Three are built in full and run right now:

Loop	What it does	Run it
1 docs and examples sync	rewrites drifted docs, runs every example to prove it works	`bash loop-packs/engineering-loops/loops/01-docs-and-examples-sync/check.sh`
19 eval suite	scores a classifier against an eval set, blocks on a regression	`cd loop-packs/ai-ml-loops/loops/19-*/example && node run-evals.js`
28 flaky test detector	runs each test 20 times, catches the unstable one	`cd loop-packs/qa-loops/loops/28-*/example && node detect-flaky.js`

Then put one in your own repo

The demos run against seeded data. The point is your repo:

./install.sh ~/code/my-app 02      put loop 2 in your project
cd ~/code/my-app
./run-loop.sh 02 --check           is there work here?

That second command is the whole idea in one line. Your first loop walks the rest.

If it will not run

What you see	What to do
`permission denied`	`chmod +x run-all-demos.sh`, or the check you ran
`node: command not found`	install Node 18 or newer, then open a new terminal
`gh: command not found`	that is only the agent step. Every check still runs.
a check exits 2	it is not wired to this folder. Run it from your project root.
findings printed, non zero exit	correct. That means "found work."

← Plain words · Contents · Next: Your first loop →

The 35 loops: Engineering · Cloud · AI and ML · QA

Reading is the slow way round. This runs all four chapters against seeded data:

git clone https://github.com/fakiolinho/From-Prompt-to-Loop.git && cd From-Prompt-to-Loop && ./run-all-demos.sh