act-post-decision

Pass

Audited by Gen Agent Trust Hub on Aug 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local scripts using bash and node (ensure-deps.sh and db.mjs). These are shared helpers located in the platform's root directory (CLAUDE_PLUGIN_ROOT) used for dependency management and database interaction.
  • [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection as it processes text from external decisions. It includes explicit instructions to treat this content as data rather than executable instructions for the agent itself.
  • Ingestion points: Neon task list (list-blocked) and Jupi decision details (get-decision) in SKILL.md.
  • Boundary markers: Includes an explicit instruction to ignore embedded commands within signal or option text.
  • Capability inventory: Can trigger tool writes via execute-action, update task statuses in Neon, mark actions as done in Jupi, and append to a rules store across ALL scripts.
  • Sanitization: No explicit sanitization logic is defined in the instructions; it relies on the boundary instruction.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 3, 2026, 12:19 PM
Security Audit — agent-trust-hub — act-post-decision