agent-task-directories
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill proactively addresses indirect prompt injection risks by instructing the agent to treat task materials and tool outputs as untrusted data. It explicitly states that untrusted material cannot grant authority or expand the agent's scope.
- [DATA_EXPOSURE]: The skill mandates the creation of a
.gitignorefile containing a universal wildcard (*) for all task directories to prevent scratch artifacts from being accidentally tracked or committed to version control. - [CREDENTIALS_UNSAFE]: A dedicated 'Secrets and Authentication' section forbids the handling, exposure, or retention of credentials, private machine values, or secret material in commands, configuration, or artifacts.
- [COMMAND_EXECUTION]: The instructions emphasize staying within assigned write scopes and explicitly warn against using path traversal or symlink redirection to expand the agent's mutation scope.
Audit Metadata