Risk-based exploratory testing

Explores high-impact workflows with targeted questions and adapts testing when unexpected behavior appears.

risk-based-exploratory-testing

See how it helps

When is it useful?

Use when scripted checks pass but new failure modes may hide in timing or state combinations.

What does it help you do?

Prioritizes business effects, changed complexity and earlier defect patterns.
Defines a focused exploration charter with inputs, roles and timing conditions.
Records discoveries and uncertainty using realistic isolated test data.

Try asking this

Explore agent launch, cancellation and restart under lost connection and changed workspace permissions, looking for inconsistent run state. Disconnect immediately after a tool starts and revoke workspace access before reconnecting. Record whether the run stops, resumes or remains uncertain using persisted evidence.

Detailed documentation

Read the skill’s original instructions, examples and checks in their authored language.

Details…Show less

Risk-Based Exploratory Testing

Automated checks verify what we already thought to script; exploratory testing (James Bach, Elisabeth Hendrickson, Michael Bolton) combines learning, test design, and execution to uncover the failure modes nobody anticipated.

Workflow

Prioritize test charters by risk and blast radius:
- Never attempt to test every theoretical permutation blindly. Rank test focus by:
  - User & business impact: Core task completion, data loss/corruption risk, permission boundaries, financial/quota accuracy.
  - Change surface & complexity: Newly modified modules, complex state machines, async coordination, and areas with prior bug history.
Formulate concise exploratory charters:
- Structure each investigation around:
  - Explore: [target feature, workflow, or component]
  - With: [specific inputs, role personas, timing/network conditions, or state variations]
  - To discover: [potential failure modes, state inconsistencies, or usability traps]
- Example charter: "Explore task creation -> agent launch -> mid-run cancellation -> restart -> result view using different workspace roles, malformed inputs, and an interrupted connection."
Apply falsification thinking (Feynman):
- Ask at each step: "What input, sequence, or timing would prove my assumption that this works is wrong?"
- Follow anomalies immediately: when a response is slightly slow, a toast flickers, or a counter looks off by one, pivot the charter to probe that seam.
Keep test data realistic and isolated:
- Use realistic domain data (multi-word strings, Unicode/accents, realistic file sizes, multi-item histories) inside isolated test workspaces without touching production records or live third-party side effects.
Know when to stop and what to automate:
- Conclude exploratory passes once the high-risk charters are covered and diminishing returns set in.
- Recommend high-value, recurring checks for permanent automated test coverage.

Included files 1

SKILL.md