context-engine
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill is designed to handle sensitive company data stored in
~/.claude/company-context.md. It mitigates exfiltration risks by implementing a mandatory anonymization protocol defined inreferences/anonymization-protocol.md. This protocol requires the agent to transform sensitive data (e.g., exact financial figures, names, and runway details) into relative ranges or generic descriptors before any data is sent to external tools, APIs, or web searches. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from local context files and user conversations, which represent an attack surface for indirect injection. The risk is addressed through structured field requirements, explicit privacy rules, and a protocol that requires user confirmation before any modifications are made to the local context file.
- Ingestion points: Reads from the
~/.claude/company-context.mdfile and processes information provided by the user during C-suite advisor sessions. - Boundary markers: The skill uses a 'Privacy Rules' section and a 'Decision Tree' within the anonymization protocol to delimit safe and unsafe data handling.
- Capability inventory: The skill facilitates local file access for reading and updating the context file.
- Sanitization: Implements systematic transformations of sensitive data points (anonymization) before they are interpolated into external requests.
Audit Metadata