using-superra
Warn
Audited by Gen Agent Trust Hub on Jul 8, 2026
Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill contains explicit instructions directing the AI to override the default system prompt and harness guidelines if they conflict with the skill's logic or user commands.
- [COMMAND_EXECUTION]: The skill requires running a local CLI tool (
./superRA/superra) for reading and editing tasks, and it initiates a background server on localhost to host a dashboard. - [EXTERNAL_DOWNLOADS]: The skill describes a setup and bootstrap process that installs custom agent configurations into the user's home directory (
~/.codex/agents/) and prepares the local environment for execution, which may involve fetching external resources. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes "task tree" files (e.g.,
task.md) which contain instructions that influence agent behavior. - Ingestion points: Task tree files located in the
superRA/directory. - Boundary markers: The skill uses a specialized reader (
superra task read) but lacks explicit instructions to ignore potentially malicious embedded commands within the tasks. - Capability inventory: Subprocess execution (CLI tool), file writing, network server management, and subagent spawning.
- Sanitization: No explicit sanitization or filtering mechanisms are documented for the data ingested from task files.
Audit Metadata