Index of every page on this site: llms.txt
All skills
raphaelsalaja avatar

/stress

@b2e4283
Run it yourself

Swaps the worst real data into a screen through its own data boundary, then reports every value that breaks it.

Use this Skill: https://skilld.dev/gh/raphaelsalaja/skills/stress

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈30 tokens for metadata: the name and description. ≈2k when used: this file. ≈2.3k more on demand in 2 files.

How this Skill uses context

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

This Skill turns off automatic use in Claude Code. Its description stays out of the automatic Skill list by default.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Stress

This skill runs a screen or component against the worst data it can receive. It maps every value the screen renders, swaps in a worst-case fixture through the same boundary as the demo data and records what breaks.

Real limits, every case rendered

Each limit comes from the validation schema, the database column or the API contract. Where none of the three sets one, the map says "no limit found" and the longest case applies. The cases in the catalog are exact strings, so paste each one as written. Run it on any screen that shows data a person or another system writes.

Longer translated and pseudo-localised interface text, other scripts, mirrored layouts and the width set belong to viewports. Which states a screen shows, such as empty or failed, belong to states. Combinations the data should never hold belong to invariants. Opening the running result belongs to verify. A break another skill owns cites that skill's id, such as viewports-no-clipped-text.

1. Map every value

List every value the screen renders. Beside each one, write its source and its limit, read from the validation schema, the database column or the API contract.

Value          Source                      Limit
title          projects.title              varchar(255), no minimum
owner name     users.name                  no limit found
owner avatar   users.avatar_url            no limit found
progress       tasks API, done / total     0 to 100, not enforced
due date       projects.due_at             timestamptz, nullable

A fixture built before the map tests only the values someone remembered, and is stress-map-every-value. Note where two limits disagree, such as an input capped at 50 characters that saves to a 255-character column. That mismatch is reported under invariants.

When the user gives read access to real data, add the longest stored value of each text field beside its limit. Query lengths only, such as SELECT max(length(title)) FROM projects, and never copy a real value into the fixture. Never write to that database. A schema limit with the real longest value left unchecked is stress-longest-real-value.

2. Change the data, never the markup

Build the worst-case fixture beside the demo data. Load it through the same boundary the demo data uses: the fixture file, the mock handler or the component's props. Text typed into the rendered markup or the DOM skips the code that formats, escapes and truncates it, and is stress-change-the-data.

Write each fixture value literally, or generate it from a fixed seed. Pin "now" to one instant, such as 2025-06-15T12:00:00Z, and set every date relative to it. Hand the screen's clock the same instant, through a now prop, the formatter's reference date or a development-only clock the switch sets. Otherwise the "Now" case reads "1 year ago". Math.random(), an unseeded generator or new Date() gives new rows on each reload, so a break cannot be shown twice. That is stress-fixed-fixture.

Add one switch that picks the data set: Demo, Worst Case, Empty, One and 1,000 Rows. It reads and writes ?data= in the URL as demo, worst, empty, one and 1000, so a reload keeps the case. Build 1,000 Rows by cycling the catalog's cases, so no two neighbours match.

The switch also carries the worst responses. Slow as ?data=slow delays every response 3 seconds and then serves the demo data. Failed as ?data=error answers every request with a 500. What the screen shows then belongs to states. A switch that can only serve good data is stress-slow-and-failed.

Build the switch with the project's own components. Render it in development only and gate the fixture loader the same way. A switch or fixture reachable in production is stress-dev-only-switch.

3. Run the catalog

Fill each mapped value with every case in catalog.md that fits its type. Add two cases per limit in the map: a value exactly at the limit and one a character past it. A case skipped because "nobody has a name that long" is stress-run-the-catalog.

4. Render at the real width

Render each case inside the component's real container at the width it ships at, such as a 280px sidebar or a three-column grid at 1440. A card checked alone on an empty full-width page is stress-real-container.

Record each break with the case, the value and what happened. A break is text past its box, an ellipsis with no way to read the rest or wrong initials. It is also markup that runs, NaN, Invalid Date, a raw float, a crash or a value lost on save.

For every case a form can edit, open the form, save without changing anything and reopen it. A value the round trip changed is a break. Screenshot every case that breaks.

5. Keep the fixture

Keep the worst-case fixture and the switch after the run, beside the demo data. When the project has component, snapshot or visual tests, add one that renders the fixture with the runner it already uses. A fixture deleted once the screen looks fine is stress-keep-the-fixture. When the project has no tests, the fixture and the switch are the regression check.

6. Report by case

Return one screenshot per broken case, then the findings in the same order. A finding with no case named, such as "long names overflow", is stress-name-the-case.

The pair

Every check is a pair. The broken line sits beside the line that replaces it.

Rule Broken Replacement
stress-map-every-value A fixture of the three values someone remembered Every rendered value listed with its source and limit
stress-longest-real-value "Titles are varchar(255)" with the stored data never measured "Titles are varchar(255), the longest stored is 212", with both lengths in the fixture
stress-change-the-data A long name typed into the DOM The name in fixtures/worst.ts, loaded through the props
stress-fixed-fixture Names from an unseeded generator, dueAt: new Date() The generator seeded with 7, now = "2025-06-15T12:00:00Z" and dueAt 3 days after now
stress-slow-and-failed A switch with Demo and Worst Case only ?data=slow holding every response 3s and ?data=error answering 500
stress-dev-only-switch ?data=worst working on the production site The switch and the fixture loader gated to development
stress-run-the-catalog "Nobody has a name that long" Christopher Alexander Montgomery III rendered in the row
stress-real-container A card checked alone at 1440 The card inside the 280px sidebar it ships in
stress-keep-the-fixture The fixture deleted after the run fixtures/worst.ts kept and rendered by the existing visual test
stress-name-the-case "Long names overflow" "At 280px, the 64-character email pushes the role off the card"

Reporting

Severity. HIGH is markup that runs, a crashed render, a value lost on save or a switch reachable in production. MEDIUM is text past its box, a clip with no way to read the rest, wrong initials, NaN, Invalid Date or a raw float. It is also a fixture typed into the markup, a catalog case skipped, a case run outside its real container or a run with no value map. So is a fixture that changes on reload, or a switch with no slow or failed response. LOW is a fixture deleted after the run, or a finding that does not name its case. It is also a limit left unmeasured against readable real data.

Format. One finding per root cause, naming the case and the width:

1. file:line - [rule-id] case, at 280px, what is wrong (measured values)

Then a Rule | Count | Severity summary. Block when any HIGH remains, Approve otherwise. Mark every case the data boundary could not carry as Not verified, such as a slow image with no way to delay the response.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alertstoday3 checks · Risk SAFE
  • Gen Agent Trust Hubtoday

    The 'stress' skill is designed for UI stress testing and includes various edge-case strings to verify application robustness. It involves reading database metadata to determine field limits and generating dynamic test data. While it includes special Unicode characters and database queries, these are integral to its intended testing functionality.

  • Sockettoday

    No alerts

  • Snyktoday

    Risk: LOW · No issues

Signed by skilld at b2e4283. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 12 hours ago.

Activeupdated yesterday
disable-model-invocation
true

README badge

README badge for raphaelsalaja/skills/stress