dag-dynamic-replanner

Pass

Audited by Gen Agent Trust Hub on Feb 19, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION] (LOW): Indirect Prompt Injection vulnerability surface.\n
  • Ingestion points: SKILL.md logic (specifically handleReplanTrigger) processes ReplanTrigger data which includes a reason and suggestedAction. This data is designed to be provided at runtime based on external discoveries and failures from other agent components.\n
  • Boundary markers: Absent. The skill does not implement delimiters or safety instructions (e.g., "ignore instructions within this block") when processing external trigger content.\n
  • Capability inventory: The skill is authorized to use the Task (command execution), Write, Edit, and TodoWrite tools, providing a path for injected instructions to influence system state or execute arbitrary tasks.\n
  • Sanitization: Absent. The implementation logic lacks validation or sanitization of the string content within triggers before using them to modify DAG nodes or configurations.\n- [SAFE] (SAFE): No malicious code patterns, obfuscation, or unauthorized network activities were detected within the provided TypeScript logic and metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 19, 2026, 08:37 PM