Three small, tested Python tools for jobs that eat hours every week: cleaning spreadsheet exports, keeping two systems in sync, and shipping releases without mistakes.
Standard library only (Python 3.10+): nothing to install, runs on Windows, macOS and Linux. 43 tests run on every push. All data in this repository is made up.
| Tool | The job | What makes it safe to rely on |
|---|---|---|
csv_report |
Merge messy CSV exports into one clean table, an Excel file and an HTML report | Flags values it can't read instead of guessing; keeps leading zeros; newer exports override older ones |
api_sync |
Push a price list from a CSV into a REST API | Dry run by default; retries with backoff; reads every write back to prove it stuck |
release_packager |
Build a release zip with checksums | Dry run by default; never overwrites a release; blocks secrets; byte-identical rebuilds |
$ python csv_report/csv_report.py "csv_report/sample_data/*.csv" --key order_id --group-by region --sum amount
read 10 row(s) from 2 file(s); 2 duplicate(s) removed; 8 clean row(s)
column types: order_id=number, order_date=date, customer=text, region=text, amount=number, zip=text
issue 2026-01_orders.csv:6 amount 'twelve' not a number (left blank)
wrote report_out/clean.csv, clean.xlsx, report.html
The two sample files are messy the way real exports are: one is UTF-8 with commas, the other Windows-1252 with semicolons and decimal commas; headers differ (Order ID vs order-id); dates come in three formats; amounts look like "$1,250.00", (20.00) and N/A; one order appears in both months.
- Reads what people actually send: detects the delimiter and encoding, normalizes headers so different column names and orders merge, understands
$1,234.50,(5.00),-$15.00,120,50and1.250,00. - Never guesses silently. A column is treated as numbers or dates when at least 80% of it parses; every value that doesn't is listed with its file and line.
- Keeps what matters:
04011stays04011(ZIP codes, IDs). With--key, a later file's row replaces an earlier one, so the newest export wins. - Three outputs:
clean.csv,clean.xlsx(real numbers and dates, bold header, filter, frozen top row) andreport.html(the screenshot above).--strictreturns exit code 1 when anything needs review, for use in scheduled jobs.
$ python api_sync/api_sync.py --base-url http://127.0.0.1:8765 --source api_sync/sample_prices.csv
plan: 1 to create, 2 to update, 1 unchanged, 1 only in the API (left alone)
+ D-400 Laptop sleeve 14 in price 19.00 stock 25
~ A-100 price: '24.99' -> '22.99'
~ B-200 stock: 80 -> 64
= C-300 only in the API, not touched
dry run: nothing was written. Run again with --apply to make 3 change(s).
$ python api_sync/api_sync.py --base-url http://127.0.0.1:8765 --source api_sync/sample_prices.csv --apply
...
applied: 3 written and verified, 0 failed, 0 read-back mismatch(es)
- Checks the whole source first. One bad row (empty SKU, a price like
abc, stock1.5, a duplicate) stops the run before anything is sent, with every problem listed. - Shows the exact changes field by field, and never deletes anything that is only in the API.
- Survives a shaky connection: network errors, timeouts and HTTP 429/5xx are retried with exponential backoff and jitter, and
Retry-Afteris respected. Retrying is safe because every write is an idempotentPUT. - Proves the result. After writing, it reads every item back. An API that answers "200 OK" but doesn't store the change is reported as a mismatch (exit code 1), not counted as done.
- Keeps the token out of sight: read from an environment variable, sent only in the
Authorizationheader, never printed or written to the--reportJSON.
Try it locally with the included demo API:
python api_sync/fake_api.py --port 8765 # terminal 1
export SYNC_API_TOKEN=local-demo # terminal 2 (PowerShell: $env:SYNC_API_TOKEN="local-demo")
python api_sync/api_sync.py --base-url http://127.0.0.1:8765 --source api_sync/sample_prices.csv
$ python release_packager/package_release.py release_packager/sample_app --name sample-app --version 1.4.0 \
--changelog release_packager/sample_app/CHANGELOG.md --go
ok version 1.4.0 is MAJOR.MINOR.PATCH
ok version is new (no earlier release before it)
ok no secret-looking files among 4 file(s)
ok CHANGELOG.md mentions 1.4.0
files 4 file(s), 229 B before compression:
sample-app/CHANGELOG.md (101 B)
...
ok verified 4 file(s) inside the zip against the source
built dist/sample-app-1.4.0.zip 730 B sha256 c44241c0...
$ python release_packager/package_release.py release_packager/sample_app --name sample-app --version 1.3.9
STOP version 1.3.9 is not higher than the latest release in dist (1.4.0)
- Dry run by default. Nothing is written until you add
--go. - Stops at the first failed check and says what to fix: version format, version not newer than the last release, the release already exists, a changelog without the version, or a file that looks like a secret (
.env, private keys,*.pem, cloud keys inside a text file). - Reproducible: sorted entries and fixed timestamps, so the same input gives a byte-identical zip and the same SHA-256.
- Verified before it counts: the zip is written under a temporary name, reopened, checked file by file against the source, and only then renamed. Writes
.sha256(works withsha256sum -c) and a JSON manifest of every file.
python -m unittest discover -s tests -v
The API tests start the demo API on 127.0.0.1 on a free port and never wait for real backoff delays. Every safety feature above has a test, and each test was checked to fail when the feature is removed.
csv_report/ csv_report.py + sample_data/
api_sync/ api_sync.py, fake_api.py (demo/test server), sample_prices.csv
release_packager/ package_release.py + sample_app/
tests/ one test file per tool
MIT. See LICENSE.
Made by Off the Clock Software. We build tools like these to order, fitted to your files, your API and your release process: offtheclocksoftware.com.
