delivering-goal

Pass

Audited by Gen Agent Trust Hub on Aug 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The subagent templates (goal-planner.md and phase-decider.md) contain instructions to execute git commands (status, diff, log, add, commit) for version control of the goal-tracking documentation. These commands are scoped to the project's goal directories and include security instructions to exclude sensitive files like .env or credentials from being staged.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests goal statements and external source descriptions to generate instructions for subagents. It uses structured templates and strict one-line handoff protocols to minimize the risk of malicious instructions influencing the execution loop.
  • [SAFE]: The skill performs local directory discovery across various hidden folders (e.g., .claude/skills/, .cursor/agents/) to locate its own subagents and project-specific skills. This is a functional requirement for cross-environment portability and does not involve unauthorized data access or network transmission.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 16, 2026, 05:49 AM
Security Audit — agent-trust-hub — delivering-goal