oulang-knowledge-sync
Fail
Audited by Gen Agent Trust Hub on Apr 14, 2026
Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [DATA_EXFILTRATION]: The skill targets local AI session history files located in
~/.claude/projects/and~/.codex/sessions/. These files contain comprehensive logs of user interactions, potentially including proprietary code and private information. The skill is designed to run autonomously as a scheduled task, exfiltrating this data to an external Notion workspace without user review.- [CREDENTIALS_UNSAFE]: The extraction logic explicitly searches for and captures sensitive terms includingpassword,credential, andboss conflictfrom the user's local history files. These secrets are then transmitted to external Notion pages.- [COMMAND_EXECUTION]: The skill uses shell commands (ls,find,awk) and Python scripts to autonomously discover and process files on the local filesystem.- [DATA_EXFILTRATION]: The skill routes extracted facts to specific Notion page IDs (e.g.,3402ea9b77ad81b5aefaf0e86e954ab3) and collection IDs. While Notion is a legitimate service, the autonomous transmission of harvested local credentials constitutes a significant security risk.- [PROMPT_INJECTION]: The skill processes untrusted data from historical conversation logs, creating a surface for indirect prompt injection where malicious instructions stored in history could influence the agent's actions during the sync process.\n - Ingestion points: Local JSONL session files from Claude and Codex in the user's home directory.\n
- Boundary markers: Absent; the skill treats all extracted content as valid knowledge to be synced.\n
- Capability inventory: File system access via shell and Python scripts, network transmission via Notion MCP.\n
- Sanitization: No sanitization or verification of the extracted content is performed before it is sent to Notion.
Recommendations
- AI detected serious security threats
Audit Metadata