agentleFS
DonaldMurilloDonaldMurillo/gofastr/.claude/skills/chaos-test/SKILL.md

chaos-test

Skill24 starsChanged 2 days agoMITView on GitHub

Live-agent chaos testing: spawn diverse persona agents against the running dev server, each with playwright browser tools and a distinct exploration mandate. Use when the user says "chaos test", "stress test the UI", "find bugs by exploring", "monkey test the site", or invokes `/chaos-test`. Personas are non-deterministic explorers, NOT chromedp scripts; they make judgment calls, follow surprising paths, and report what they tried. Triggers on phrases like "chaos monkey", "exploratory testing", "stress the UI", "agentic testing".

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 4 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When this skill is invoked, you are the orchestrator.

    line 8
  2. Distinctive:

    Do NOT use this skill for:

    line 20
  3. Distinctive:

    Before spawning personas, confirm:

    line 27
  4. Very specific:

    Check with curl -s -o /dev/null -w "%{http_code}\n" <url>.

    line 30
  5. Very specific:

    A run directory exists: Create /tmp/gofastr-chaos/<ISO-timestamp>/.

    line 32
  6. Distinctive:

    Wait for all 7 to return.

    line 62
  7. Very specific:

    Read every report file. Collate into a single triage doc at /tmp/gofastr-chaos/<run>/summary.md with sections:

    line 63
  8. Very specific:

    If you have mcp__pi__* available and the orchestration phase needs:

    line 77
  9. Very specific:

    When the user invoked the skill, say in one line how many agents run and for how long (seven in parallel for about ten minutes by default.

    line 88

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Chaos-test: live exploratory testing via diverse persona agents

When this skill is invoked, you are the orchestrator. The skill replaces deterministic chromedp tests with live agents who actually drive a browser and make judgment calls.

When to use

Invoke this skill when:

  • The user wants stress / chaos / exploratory testing of a running UI
  • Deterministic tests miss real-world weirdness (rage clicks, weird inputs, edge-case keyboard flows, slow networks)
  • You've shipped a UI change and want to surface what a curious user finds

Do NOT use this skill for:

  • Adding regression tests (use chromedp in examples/site/e2e_*.go)
  • Code review (use Agent with general-purpose, or a UX-review agent if your environment provides one)
  • Static analysis of the codebase

Pre-flight

Before spawning personas, confirm:

  1. The dev server is running at the target URL (default http://localhost:8082). Check with curl -s -o /dev/null -w "%{http_code}\n" <url>. If not 2xx, ask the user to run ./scripts/dev-watch.sh (or override target via args).
  2. A run directory exists. Create /tmp/gofastr-chaos/<ISO-timestamp>/. Each persona writes a <persona>.md report there.

Personas (7)

Spawn in parallel via Agent tool with subagent_type=chaos-<persona-name>:

PersonaMandate
chaos-rage-clickerSpam clicks, double-clicks, drags. Surface race conditions and stuck UI states.
chaos-keyboard-sherlockTab/Shift-Tab/Enter/Esc only. Find focus traps, broken focus order, invisible focus.
chaos-mobile-thumb320/375/414 viewports. Hamburger, horizontal scroll, tap-target overlap, touch gestures.
chaos-form-vandalSubmit forms with emoji, RTL, 10MB strings, <script>, control chars, malformed dates.
chaos-console-auditorVisit every reachable route. Capture console.error/warn, uncaught errors, CSP violations, 4xx/5xx network.
chaos-network-pessimistThrottle network/CPU, drop requests, go offline. Check graceful degradation.
chaos-screen-readerUse accessibility tree. Validate aria-live announcements, landmarks, focus announcements.

Each agent has playwright browser tools and writes its findings to /tmp/gofastr-chaos/<run>/<persona>.md in the structured format defined in its own agent file.

Spawning protocol

  1. Confirm dev server is reachable.
  2. Make the run directory.
  3. Spawn all 7 personas in parallel via a single message with 7 Agent tool-use blocks. The dev server handles the concurrency. Each agent gets:
    • The target URL
    • The path to its report file
    • A 10-minute time budget (the agent decides when to stop)
  4. Wait for all 7 to return.
  5. Read every report file. Collate into a single triage doc at /tmp/gofastr-chaos/<run>/summary.md with sections:
    • Critical findings: bugs that crash, lose data, fail WCAG-A, or render content unreadable. Cite which persona found each + reproduction steps.
    • Quality findings: UX rough edges, missing affordances, perf glitches.
    • What worked: paths every persona reported as smooth (lets you see where the code is genuinely solid, not just where bugs are).
    • Inter-persona corroboration: findings reported by ≥2 personas get promoted (high-signal, multi-angle confirmation).
  6. Present the summary to the user with a recommendation: fix all, triage tier-1 only, or accept as-is. The user picks.

Pi delegation

If you have mcp__pi__* available and the orchestration phase needs:

  • Cross-report deduplication (5 personas all flagged a contrast issue)
  • Severity classification of a long list
  • Summarization of repetitive findings

…delegate via mcp__pi__pi_delegate or mcp__pi__pi_swarm. The browser-driving work stays with Claude Agents (playwright is a Claude MCP).

Cost note

Seven parallel playwright sessions for ~5–10 min each = serious token spend. When the user invoked the skill, say in one line how many agents run and for how long (seven in parallel for about ten minutes by default; fewer or longer with --persona or --budget) and that it uses a lot of tokens, then start. Ask before starting only when running it was your idea.

Args

  • --target <url>: override the default http://localhost:8082
  • --persona <name>: run just one persona instead of all seven
  • --budget <min>: per-agent time budget in minutes (default 10)

What success looks like

A summary.md that:

  1. Is short enough to read in a few minutes: lead each tier with the findings that matter most and group repeats of one root cause
  2. Names specific reproduction steps for every critical finding
  3. Cites which persona surfaced each issue (so you can ask follow-ups)
  4. Distinguishes "X is broken" from "X is a UX rough edge"
  5. Names the things that worked, not just the bugs

Then leave the run directory in place; it's a paper trail.

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.