rant-coding
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructs the agent to treat transformed user input as the authoritative task and execute it immediately. This creates a surface for indirect prompt injection where malicious instructions hidden in a technical prompt could be processed and then executed by the agent.
- Ingestion points: User-supplied prompts provided via the
/rant-codingcommand are ingested into the agent's context for transformation. - Boundary markers: There are no explicit delimiters (like XML tags or unique string markers) defined to separate the untrusted input from the skill's own logic, nor are there instructions for the agent to ignore embedded commands within the source text.
- Capability inventory: Upon transformation, the agent uses its full suite of default tools and capabilities to carry out the resulting task.
- Sanitization: The skill does not perform programmatic sanitization or validation of the input prompt before rewriting and execution, relying solely on the base model's underlying safety guardrails.
Audit Metadata