True Seeing

PassAudited by ClawScan on May 2, 2026.

Overview

This is a coherent instruction-only fact-checking skill that uses web search and task-scoped sub-agents in a disclosed, bounded way.

This skill appears safe to install as an instruction-only fact-checker. Before using it, remember that it works by turning article facts into web searches and fetching public pages, so do not use it on confidential drafts unless that data flow is acceptable. Always review the cited sources and approve replacements manually.

Findings (3)

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.

What this means

The agent may make multiple web searches and fetch web pages while checking an article.

Why it was flagged

The skill directs the agent to use web search and web fetch tools, including parallel sub-agents, which is a meaningful tool-use capability. It is disclosed and directly aligned with the fact-checking purpose.

Skill content
每个 sub-agent 执行 `web_search`,取前 3 条结果用 `web_fetch` 抓取内容。
Recommendation

Use it for articles where web-based verification is appropriate, and review the generated source links before accepting corrections.

What this means

Incorrect or misleading web pages could influence the suggested corrections.

Why it was flagged

The skill bases verification decisions on retrieved web snippets. This is expected for fact-checking, but web content can be wrong, stale, or adversarially written, so the cited evidence should not be blindly trusted.

Skill content
`source_snippet` 必须包含足以证明分歧的关键句子
Recommendation

Check the cited URLs and snippets, especially for important legal, medical, financial, or reputational claims.

What this means

Private details from an article could be included in search queries or sub-agent tasks during verification.

Why it was flagged

The skill sends fact-checking work to parallel sub-agents. This is disclosed and bounded, but it means article-derived factual claims may be distributed across agent sessions and external search/fetch operations.

Skill content
使用 `sessions_spawn` 启动 sub-agent 并行搜索
Recommendation

Avoid using it on confidential unpublished drafts unless you are comfortable with article facts being used in web searches.