oulang-knowledge-capture

Warn

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill instructions specifically command the agent to extract and store 'Credentials, payment accounts, API endpoints' from conversation history into Notion. This practice leads to the exposure of secrets in a non-secure environment.
  • [DATA_EXFILTRATION]: The workflow is designed to harvest and move highly sensitive corporate data, including 'Business numbers (revenue, users, traffic)', 'Contract, IP, payments owed', and 'Boss/stakeholder quotes', to an external Notion platform.
  • [PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection (Category 8) due to its handling of untrusted conversation data. 1. Ingestion points: The entire 'CURRENT conversation' history is used as the primary data source (SKILL.md). 2. Boundary markers: No delimiters or ignore-instructions are present to separate user data from agent instructions. 3. Capability inventory: The skill utilizes the Notion MCP toolset with capabilities to fetch pages, update content, and create database entries. 4. Sanitization: No sanitization or validation logic is defined; the skill explicitly requires 'VERBATIM' extraction of facts.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 12, 2026, 11:20 PM