smart-compaction

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the duoduo CLI tool to get and set session configurations (duoduo session config). These operations are local to the platform's environment and are used to manage the 'Smart Compaction' feature.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill mentions accessing pre-compact transcripts via file paths provided in system notices (<smart-compact-notice>). This is a documented feature for recovering conversation history that was removed from the active context window and does not involve exfiltration to external domains.
  • [REMOTE_CODE_EXECUTION]: The skill mentions that Grok-based sessions execute _x.ai/compact_conversation. This appears to be a platform-internal command or endpoint for that specific backend and not an arbitrary code execution vector for the agent.
  • [INDIRECT_PROMPT_INJECTION]: The skill reacts to a <smart-compact-notice> block. This block is defined as being injected by the platform 'kernel' (system-level) rather than user-generated content, mitigating the risk of injection from untrusted sources.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 02:02 PM
Security Audit — agent-trust-hub — smart-compaction