Run any Skill in Manus with one click

prompt-stress-test

Stars0

Forks0

UpdatedApril 13, 2026 at 01:15

Use this skill when the user wants to deliberately break, stress test, or find failure points in an existing Claude prompt or skill before shipping it. Triggers on phrases like "stress test this prompt", "where will this break", "find edge cases", "adversarial test", "QA this skill", or any request to probe a prompt with hostile/ambiguous inputs and get a failure report BEFORE making fixes. Do NOT use when the user already knows the failures and wants fixes (use prompt-refinement), or when they want a holistic reliability score without adversarial testing (use prompt-reliability-score).

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

composite

composite/enhanced-skills

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software Quality Assurance Analysts and TestersComputer and Mathematical Occupations·SOC 15-1253

SKILL.md

readonly

name

prompt-stress-test

description

Prompt Stress Test

Break prompts on purpose. Report failures. Do not fix.

Inputs required

The current prompt or skill (verbatim)
The intended use case (one sentence)

If either is missing, ask for it before proceeding.

Procedure

Generate 10 adversarial inputs across these classes: ambiguous phrasing, off-topic requests, edge-of-scope inputs, contradictory instructions, missing context, hostile tone, extreme length (too short/too long), format violations, multilingual or encoding quirks, and inputs that exploit stated constraints.
Run each input against the prompt mentally and record the actual failure mode.
Score each failure: 1 (minor drift) / 2 (inconsistent output) / 3 (wrong answer) / 4 (complete breakdown).
Rank failures by severity × likelihood.
Deliver the report. Do not propose fixes unless explicitly asked.

Rules

Adversarial inputs must be realistic for the stated use case. No absurd edge cases.
Every failure must include: the input, the observed break, the severity score, and a one-line root cause hypothesis.
Report before touching the prompt. Fixing is out of scope for this skill.
If fewer than 3 failures are found across 10 inputs, state that explicitly rather than inventing failures.

Output format

Adversarial Input Set — numbered list, 10 items
Failure Log — table: input # | failure mode | severity | root cause hypothesis
Severity Ranking — failures sorted high → low
Stress Test Score — X/40, with a one-sentence verdict

More from this repository

same repository

prompt-refinement

composite/enhanced-skills

Use this skill when the user has an existing Claude prompt and wants it rewritten to fix known problems — failure patterns, vague instructions, format drift, or bloat. Triggers on phrases like "fix this prompt", "tighten this", "my prompt keeps failing", "the output format keeps drifting", "replace vague instructions", "cut the filler", or any request to modify an existing prompt while preserving its intent. Handles four refinement modes: failure repair, constraint tightening, output format locking, and bloat removal. Do NOT use for building a new skill from scratch (use skill-construction) or for evaluation only (use prompt-stress-test or prompt-reliability-score).

2026-04-130

prompt-reliability-score

composite/enhanced-skills

Use this skill when the user wants a pre-launch reliability score or readiness audit of a Claude prompt across multiple quality dimensions, without adversarial testing. Triggers on phrases like "score this prompt", "is this ready to ship", "rate my prompt", "reliability audit", "launch readiness check", or requests for a dimensional breakdown of prompt quality. Do NOT use when the user wants adversarial failure-finding (use prompt-stress-test) or when they want the prompt rewritten (use prompt-refinement).

2026-04-130

skill-construction

composite/enhanced-skills

Use this skill when the user wants to build a new Claude skill from scratch, expand a single prompt into a multi-phase skill system, or add a self-improvement feedback loop to an existing skill. Triggers on phrases like "build a skill for", "design a prompt that won't break", "turn this prompt into a full workflow", "make this skill self-improving", "production-ready skill for X", or any request for new skill architecture (not just editing an existing prompt). Handles three construction modes: greenfield design with failure prevention, single-prompt expansion into a phased system, and feedback loop integration. Do NOT use for editing an existing prompt in place (use prompt-refinement) or for evaluation (use prompt-stress-test / prompt-reliability-score).

2026-04-130

name

prompt-stress-test

description

Prompt Stress Test

Break prompts on purpose. Report failures. Do not fix.

Inputs required

The current prompt or skill (verbatim)
The intended use case (one sentence)

If either is missing, ask for it before proceeding.

Procedure

Generate 10 adversarial inputs across these classes: ambiguous phrasing, off-topic requests, edge-of-scope inputs, contradictory instructions, missing context, hostile tone, extreme length (too short/too long), format violations, multilingual or encoding quirks, and inputs that exploit stated constraints.
Run each input against the prompt mentally and record the actual failure mode.
Score each failure: 1 (minor drift) / 2 (inconsistent output) / 3 (wrong answer) / 4 (complete breakdown).
Rank failures by severity × likelihood.
Deliver the report. Do not propose fixes unless explicitly asked.

Rules

Adversarial inputs must be realistic for the stated use case. No absurd edge cases.
Every failure must include: the input, the observed break, the severity score, and a one-line root cause hypothesis.
Report before touching the prompt. Fixing is out of scope for this skill.
If fewer than 3 failures are found across 10 inputs, state that explicitly rather than inventing failures.

Output format

Adversarial Input Set — numbered list, 10 items
Failure Log — table: input # | failure mode | severity | root cause hypothesis
Severity Ranking — failures sorted high → low
Stress Test Score — X/40, with a one-sentence verdict