wiki-markdown
Fail
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses the
webfetchtool to retrieve a remote file fromhttps://raw.githubusercontent.com/kepano/obsidian-skills/main/skills/obsidian-markdown/SKILL.mdwhenever the skill is activated. - [REMOTE_CODE_EXECUTION]: The instruction to "Follow [the downloaded file] as the baseline standard" constitutes remote instruction injection. In the context of an LLM agent, fetching and obeying instructions from a third-party, non-verified repository is functionally equivalent to remote script execution (e.g.,
curl | bash), as it allows an external actor to control the agent's behavior and logic without local review or verification.
Recommendations
- AI detected serious security threats
Audit Metadata