oc-toolsmith
Warn
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill instructions direct the agent to author new scripts (Python, Bash, or Node.js) and execute them via a self-test mechanism during registration (
oc toolbox new --selftest ...). This involves runtime code generation and execution. - [PERSISTENCE]: A core feature of this skill is the registration of new tools into a persistent, git-tracked
toolbox/directory. These tools remain available for all future agent sessions, creating a mechanism for executable code persistence. - [INDIRECT_PROMPT_INJECTION]: The agent authors tools based on task requirements, which creates a surface where untrusted data or malicious task instructions could influence the agent to create tools with dangerous side effects. The skill includes safety rules (e.g., no network access, no deleting user files), but these are natural language instructions rather than technical enforcement.
Audit Metadata