agentleFS
th3r-devth3r-dev/th3-skills/skills/qa/export-evidence/SKILL.md

qa-export-evidence

Skill2 starsChanged 2 days agoMITView on GitHub

Build QA test evidence for a web app's export buttons (CSV / Excel / PDF / Print) and turn it into an .xlsx test report: walk each page on a signed-in CDP Chrome, screenshot every step (page → filter → result → export button → queued alert → queue modal → the downloaded file opened), verify the file with a real parser, then write a workbook with a Summary sheet and one sheet per case (B2 title · B4 Test Result · B5 Test Environment · B6 Test Date · B7 Actual Result · captioned screenshots in a 2-column grid). Use WHENEVER the work is "test evidence for export", "test the Export Excel/PDF/CSV buttons", "QA test report xlsx", "ทดสอบปุ่ม Export", "หลักฐานการทดสอบ export", "make_qa_report", or a request to rebuild a QA report workbook from an evidence folder.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Build the report and hand it over: If it goes to a shared drive, upload through the drive's own file chooser, then confirm the file is in …

    line 49
  2. Distinctive:

    Pick a date range that returns rows: before blaming the export.

    line 78
  3. Very specific:

    One collector at a time: Every CDP client sets Chrome's download behaviour, so a second client sends the first one's files to ~/Downloads …

    line 83
  4. Very specific:

    Put each FAIL's cause into cases.json "note", then run collect_export.py … --index-only: that rewrites index.csv without opening a browser, …

    line 89

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

QA evidence for export buttons

In a list of pages × formats on one environment · Out <run>/ with cases.json, evidence/ (one folder per case: steps.csv, run.json, the downloaded file, screenshots/), and the report .xlsx.

RUN=qa-reports/<run>                      # change this line only
S=skills/qa/export-evidence/scripts

# a Chrome with remote debugging, signed in by a person:
#   chrome --remote-debugging-port=9222 --user-data-dir=~/.cache/qa-chrome [--headless=new]
uv run $S/collect_export.py $RUN/cases.json $RUN/evidence                 # all cases
uv run $S/collect_export.py $RUN/cases.json $RUN/evidence <slug>          # re-run one; the others stay
uv run $S/make_qa_report.py $RUN/evidence --out $RUN/<name>.xlsx \
    --title "<Summary title>" --env UAT --date "NOV 9, 2026" \
    [--only-sheets "reports export excel"] [--labels references/labels-th.json]

Both scripts only read and write inside the folder they run from (cases.json, the evidence dir, --out, --labels); run them from the repo root.

  1. Inventory from source, not from clicking around. The app's menu / route config plus the page templates show which pages carry an export bar. Write the list down with the kind of export each one uses (see below) — the page set changes between releases, so re-check before each run.
  2. Write cases.json. One case per page × format. sheet names the group a case belongs to (--only-sheets splits one run into several workbooks). references/cases-example.json has the shape; references/ui-th.json shows the ui block for a Thai UI (SweetAlert + Bootstrap modal defaults are built in).
  3. Sign in first. A person signs the CDP profile into the app. An agent never types a password into a login form unless the owner authorised that exact run — and then it reads the value from a local, gitignored config and never prints, logs or screenshots it.
  4. Collect, then read every FAIL. The cause sits in run.json → net (the enqueue status and body, the queue row text, the export GET error). Fix what is on our side first: the date range, the selector, waits, row selection. Then re-run that slug. If the cause is a real app bug, the case stays FAIL with that evidence; nobody patches the app to make the test pass.
  5. Build the report and hand it over. If it goes to a shared drive, upload through the drive's own file chooser, then confirm the file is in the folder at the same size.

Evidence contract (what the builder reads)

filecolumns
evidence/index.csvno,title,folder,result required · sheet (group) · actual (B7 text) · any other column shows on Summary
evidence/environment.csvkey,value → Summary header block
evidence/<folder>/steps.csvstep,action,result,evidence (+ anything) · evidence = PNG path relative to <folder>

The collector writes all three, so a hand-made run (manual test, another tool) only has to follow the same columns to get the same workbook.

Two export mechanisms that look identical on screen

  • Queued: the click sends POST …<ui.enqueue_path> with the report type and format → an alert ("queued") → a queue modal → download from the queue row once it says success. Each export leaves a job in the queue, which is the only side effect. kind: auto detects it.
  • Direct: a synchronous blob GET <url><ui.export_path>?format=…. PDF is often labelled "Print" here, not "Export PDF" — the default button regex covers both.
  • Viewer (kind: stimulsoft): a row button opens a report viewer; its Save menu → a PDF item → an export-settings dialog → OK is the download. Set open to the row button text.

Traps

  • An empty table has three indistinguishable causes — genuinely empty, a scope clamp returning 200 with zero rows, or a 403. Never record "no data" as a pass without a row count on screen (ui.rows_regex picks it up from the page text).
  • Pick a date range that returns rows before blaming the export; test data is sparse.
  • Prod safety. Click only the filter, the query button, row checkboxes, the export button, the alert's OK and the queue download. Never save, submit, confirm or delete. Never press Enter outside the date inputs. Prod downloads hold real data: they stay local, only the workbook is shared, and ~/Downloads is cleaned afterwards.
  • One collector at a time. Every CDP client sets Chrome's download behaviour, so a second client sends the first one's files to ~/Downloads and leaves it a 0-byte copy.
  • "not successful" contains "successful". The queue status check looks for the failure words first (ui.failure_words), in any language.
  • Large tables need a narrow range. Too wide a range returns a "too much data" 400, and a PDF of thousands of rows can stay at "PROCESSING 100%" forever while Excel works. Record it as a finding.
  • Put each FAIL's cause into cases.json "note", then run collect_export.py … --index-only: that rewrites index.csv without opening a browser, and the note becomes the case's Actual Result text.
  • Screenshots are embedded in the .xlsx; gitignore evidence/**/screenshots/ and re-run the collector before rebuilding from a fresh clone.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.