Scientific Inquiry
PassAudited by ClawScan on May 11, 2026.
Overview
This instruction-only research skill is generally coherent and benign, but users should notice that it may use web/curl-based searches and can modify its own skill instructions only if explicitly asked.
This skill appears suitable for evidence-based research and includes a helpful confirmation step before searching. Be aware that it may query external websites after you approve a research plan, and do not authorize self-modification unless you want its future behavior changed.
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 ask the skill to modify itself, those changes could affect how it behaves in later research tasks.
Self-modification can persistently change agent behavior. The artifact discloses it and says it is only user-commanded, so this is a noteworthy permission-sensitive feature rather than evidence of malicious behavior.
This skill uses self-modification (via `skill_manage`) but ONLY when the user explicitly commands it.
Only approve self-modification when you intentionally want the skill changed, and review any proposed edits before applying them.
The agent may make outbound web requests to search engines or other public sources while carrying out confirmed research.
The skill instructs the agent to use a raw command-line network tool as a fallback search method. This is aligned with the research purpose and gated by the earlier user-confirmation step, but it is still broader than a dedicated search API.
If web_search fails — use `curl -sL` to Google/Bing/DuckDuckGo; distinguish CAPTCHA from timeout
Use the skill for public research tasks, avoid including secrets in research queries, and review command execution if your environment prompts for approval.