Skills
skills/aradotso/ai-agent-skills/agentic-context-engine/Gen Agent Trust Hub

agentic-context-engine

Warn

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The framework features a 'Recursive Reflector' that writes and executes Python code at runtime to analyze agent traces. This dynamic code execution could be compromised if processing malicious or untrusted input traces.
  • [COMMAND_EXECUTION]: The skill documentation includes examples using subprocess.run to call system CLI tools like ace config and ace validate for troubleshooting and configuration management.
  • [DATA_EXFILTRATION]: The skill includes functionality to upload execution traces to the kayba.ai remote service via the kayba upload-traces command. Users should review traces for sensitive data before uploading.
  • [EXTERNAL_DOWNLOADS]: The skill instructs users to install the ace-framework package and provides a command (kayba install-prompt) to download and install prompts/configurations from a remote repository.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing execution traces. 1. Ingestion points: learn_from_traces and analyzer.analyze. 2. Boundary markers: Absent. 3. Capability inventory: Dynamic code execution via Reflector, file system access via agent.save, and network operations via kayba commands. 4. Sanitization: No sanitization or validation of trace content is mentioned.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 16, 2026, 10:20 AM
Security Audit — agent-trust-hub — agentic-context-engine