error-recovery
Pass
Audited by Gen Agent Trust Hub on Apr 18, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The recovery playbook includes a strategy to "Rephrase prompt" and "Retry without triggering content filter" when encountering provider errors. This is an explicit instruction to evade safety guardrails.- [COMMAND_EXECUTION]: The skill utilizes shell commands like
mkdir -pwithin its file system recovery playbook to automatically create missing directories.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted error data from external APIs and file system operations without sanitization. - Ingestion points: Error messages and outputs from AI providers, network requests, and file system errors.
- Boundary markers: Absent; error classification logic relies on raw input strings.
- Capability inventory: File read/write permissions, shell command execution (
mkdir), and the ability to modify and re-execute prompts. - Sanitization: No sanitization is mentioned for processing error strings or logging data to the audit file.
Audit Metadata