goal-creator
Pass
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes a local utility script,
scripts/goal-lint.mjs, which is called by the agent to perform deterministic checks on generated goals. This script is benign, performing regex-based text analysis without any network access or unsafe file operations.\n- [PROMPT_INJECTION]: The skill processes user-supplied objective descriptions to generate goal contracts, which represents an ingestion surface for untrusted data. The skill mitigates risks associated with indirect prompt injection through explicit 'Anti-Gaming Defaults' and 'Risk Tiers' that instruct the agent to maintain safety boundaries, verify results independently, and ignore instructions that conflict with the primary objective.
Audit Metadata