Ai Agent Evaluator
PassAudited by ClawScan on May 15, 2026.
Overview
This instruction-only skill appears purpose-aligned for evaluating AI agents, with the main caution that users may share sensitive logs or conversations during evaluation.
This appears safe to use as an advisory evaluation assistant. Before pasting logs, transcripts, or customer conversations, remove personal data, secrets, credentials, and internal prompts. Also note that the package has limited source metadata, although it contains no executable code.
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 you paste real logs or conversations, sensitive information could enter the model context during the evaluation.
The skill expects users to provide agent logs, failed transcripts, or sample conversations. That is aligned with evaluation work, but those materials can contain customer data, secrets, or adversarial prompt content.
"Input": Agent logs, failed task transcripts ... "share 10 sample conversations (anonymized)"
Only share selected, relevant examples; redact secrets, personal data, customer identifiers, and internal system prompts before use.
Users have less information to verify the publisher, source history, or exact packaged version.
The registry metadata provides no source or homepage, and the SKILL.md front matter declares a different version, "3.0.0". This is a provenance/packaging clarity issue, though impact is limited because no code or install steps are present.
Source: unknown; Homepage: none; Registry metadata version: 1.0.1
Treat this as a documentation/provenance note; prefer skills with clear source links and consistent version metadata when that matters for your workflow.