tech-article-reproducibility
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [METADATA_POISONING]: The skill contains hardcoded absolute file paths pointing to a specific user's home directory (
/Users/mz/.claude/skills/tech-article-reproducibility/SKILL.md) in the subagent dispatch templates for both English and Japanese versions. This exposes the author's local username and may cause the skill to fail or behave unexpectedly on systems with different directory structures. - [INDIRECT_PROMPT_INJECTION]: The skill processes technical article drafts which represent an untrusted data source that could contain embedded instructions.
- Ingestion points: The article content is loaded into the subagent prompt via the
<path to the article file>placeholder. - Boundary markers: The dispatch template lacks protective delimiters or explicit instructions for the subagent to disregard commands found within the target article.
- Capability inventory: The skill instructs the subagent to use
WebFetchfor link verification, creating a network-based attack vector for injected content. - Sanitization: There is no evidence of content sanitization or validation before the external article is analyzed.
Audit Metadata