Audit Arithmetic
Testing & Qualityaudit-arithmeticAudit Arithmetic is an Agent Skill for testing & quality published in the ben-manes/caffeine repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Audit for arithmetic and boundary bugs (overflow, off-by-one, drift)
QUICK START
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Prompt to paste
Inspect the Agent Skill at https://github.com/ben-manes/caffeine/blob/HEAD/.claude/skills/audit-arithmetic/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/audit-arithmetic, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
Assume there is at least one arithmetic or boundary bug in this code.
Audit for:
- Off-by-one errors
- Incorrect inequality direction (< vs <=, > vs >=)
- Overflow / underflow (int, long, specifically nanoTime arithmetic)
- Implicit narrowing or widening conversion
- Incorrect rounding or truncation
- Sign errors
- Sentinel value collisions (can a valid value equal a sentinel?)
- Counter drift (can a counter diverge from the quantity it tracks over a long-running cache?)
- Timer wheel bucket arithmetic (modular arithmetic, wrap-around)
- Frequency sketch counter saturation and reset arithmetic
- Hill climber / adaptive sizing: window vs main partition adjustment (percentage calculations, rounding, overshoot, oscillation)
- FrequencySketch table size calculation (ensureCapacity rounding to power of two, interaction with maximumSize)
- Weight delta computation in UpdateTask (sign conventions, can intermediate sums overflow when weights are near Integer.MAX_VALUE?)
For each suspected issue:
- Provide a concrete input (specific key, value, weight, time values)
- Show step-by-step evaluation through the code
- Show expected vs actual behavior
- If it requires billions of operations, state the count needed
Ignore style and performance. Only report high-confidence defects supported by evaluation traces.