oh-my-wiki
Fail
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill includes functionality to automatically install system-level software using elevated privileges. In
scripts/viewers/obsidian.py, the_install_debfunction downloads an Obsidian.debpackage from GitHub and executes it viasudo apt install -y. While this is intended for user convenience during setup, the use ofsudoto install remote binaries is a high-risk pattern. - [REMOTE_CODE_EXECUTION]: The skill facilitates the download and execution of external code and binaries. It fetches release assets from the Obsidian GitHub repository and executes them. Furthermore, the installation scripts
bin/omw-install.shandbin/ensure-cli.shdownload the skill repository and perform installations viapiporpipx, which involves executing remotely sourced code. - [DYNAMIC_EXECUTION]: The skill uses dynamic module loading via
__import__inscripts/recall.pyto import standard libraries likere,json, andpathlibat runtime. While these specific instances appear to be shortcuts for standard imports, the use of dynamic loading can complicate security auditing and potentially be misused if combined with untrusted input. - [COMMAND_EXECUTION]: Extensive use of
subprocess.runandsubprocess.Popenis found across multiple scripts includingpersona_run.py,embed_install.py,adapters.py,fetch_youtube.py, andbackends.py. These calls are used to interact with various AI backends (claude,codex,gemini), manage local embedding models, and execute system utilities likeyt-dlpor Playwright. - [EXTERNAL_DOWNLOADS]: The skill performs numerous external network requests to fetch web content, YouTube transcripts, and machine learning models. These operations are conducted through
urllib, Playwright, and third-party cloud scrapers. The skill includes defensive measures such as an SSRF guard inscripts/urls.pyand a safe redirect handler inscripts/fetch.pyto mitigate risks associated with external data ingestion. - [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it ingests untrusted data from URLs, RSS feeds, and external imports (Notion/folders). To mitigate this, the skill implements a 'propose → confirm → execute' model, redacts secrets from session captures in
scripts/session_capture.py, and frames recalled context as escaped JSON data to prevent instruction leakage.
Recommendations
- HIGH: Downloads and executes remote code from: https://api.github.com/repos/obsidianmd/obsidian-releases/releases/latest, unknown (check file) - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata