Skylv Prompt Optimization Tool
PassAudited by ClawScan on May 3, 2026.
Overview
This instruction-only prompt optimization skill is broadly coherent and does not request credentials or local access, but users should notice its suggested large A/B tests and minor metadata/version mismatch.
This appears safe to install as an instruction-only prompt optimizer. Before using its A/B testing features, set explicit limits on token budget, number of test runs, duration, and whether results are real or simulated.
Findings (2)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If followed literally, optimization could consume many model calls or tokens.
The skill’s A/B testing guidance suggests a large number of prompt test runs over a multi-day period. This is aligned with prompt optimization, but users should explicitly control cost, duration, and approval before running such tests.
sample_size: 每个变体 100+ 次测试 ... duration: 7-14 天
Ask the agent to simulate first or set a small test count, budget, and stop condition before running real A/B tests.
The listed package identity is slightly inconsistent, which may make provenance harder to verify.
The registry metadata differs from the SKILL.md frontmatter version and slug. Because this is instruction-only and contains no runnable code, this is a provenance/packaging note rather than a security concern.
Registry metadata: Version: 1.0.0 ... SKILL.md: version: 1.0.2; slug: skylv-ai-prompt-optimizer
Confirm the skill name, version, and publisher before relying on it in important workflows.