decision-history
Warn
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection] (MEDIUM): The skill parses untrusted content from
CHANGELOG.mdand coordination logs using theparseChangelogandaggregateDecisionsfunctions. Since this data is presented back to the agent to help it 'understand rationale,' malicious text in the history could influence the agent's future behavior or reasoning. - [Command Execution] (LOW): The skill executes local Node.js scripts (e.g.,
hooks/bin/decision-history.mjs) to perform its tasks. This is standard behavior for CLI-based skills but assumes the project's internalhooksdirectory is secure and not tampered with.
Audit Metadata