context-engine

Pass

Audited by Gen Agent Trust Hub on Apr 5, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a 'Context Enrichment' mechanism that persists conversation data into a local file, creating an indirect prompt injection vulnerability.\n
  • Ingestion points: Conversation history where the agent 'learns' new priorities, constraints, or data (SKILL.md).\n
  • Boundary markers: None; there are no delimiters or warnings used when appending new information to the context file.\n
  • Capability inventory: File read and write access to ~/.claude/company-context.md (SKILL.md).\n
  • Sanitization: None; while the skill references an external anonymization protocol for external calls, no sanitization is performed on data written to the persistent local file.\n- [DATA_EXFILTRATION]: The skill manages a file containing highly sensitive business data such as revenue, runway, and employee names. While it contains natural language instructions for privacy, these do not provide a technical guarantee against exfiltration or unauthorized access.\n- [COMMAND_EXECUTION]: The skill requires the agent to perform file system operations (reading and writing to ~/.claude/company-context.md) which could be exploited to manipulate local files or gain persistence if the agent's logic is subverted.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 5, 2026, 08:06 PM
Security Audit — agent-trust-hub — context-engine