Audit Serialization
Testing & Qualityaudit-serializationAudit Serialization 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 serialization proxy correctness and round-trip safety
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.
Inspect the Agent Skill at https://github.com/ben-manes/caffeine/blob/HEAD/.claude/skills/audit-serialization/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/audit-serialization, 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.
Audit the serialization/deserialization behavior of the cache.
-
Proxy completeness: Does the proxy capture ALL configuration? Check: max size/weight, expiration, key/value strength, weigher, loader, removal/eviction listener, ticker, scheduler, executor, initial capacity.
-
State transfer: Are entries serialized or only configuration? If entries: are expired entries filtered? async values handled? weak/soft refs dereferenced? If not: is this documented?
-
Deserialized consistency: Correct initial state for frequency sketch, timer wheel, drain status, deques, weight counters? Correct node types?
-
Cross-version compatibility: serialVersionUID present? Can older proxies deserialize in newer versions? Default values correct for missing fields?
-
Security: Can crafted serialized form create inconsistent state? Are inputs validated? Is the proxy pattern correctly implemented?
-
Edge cases:
- Serializing during in-flight operations
- Serializing with pending removal notifications
- LoadingCache with non-serializable CacheLoader
- AsyncCache with incomplete futures
For each defect: state the field/behavior affected, before/after across round-trip, severity (data loss, incorrect behavior, security risk).