chat-history-convention
Pass
Audited by Gen Agent Trust Hub on Apr 4, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill implements a logging mechanism that writes raw user input directly to a file (
.chat-history/user-messages.log). This creates a vulnerability surface for indirect prompt injection where malicious instructions stored in the log could be executed if the agent later reads the file context. - Ingestion points: The
scripts/append-user-message.ps1script takes user input via the$Messageparameter. - Boundary markers: While entries use timestamps and a
---separator, the user message content is not escaped or delimited from the rest of the log. - Capability inventory: The skill uses
Add-ContentandNew-Itemwithin a PowerShell script to manage log files. - Sanitization: There is no evidence of message sanitization or validation before the content is appended to the log.
- [COMMAND_EXECUTION]: The skill includes and utilizes a PowerShell script (
scripts/append-user-message.ps1) to perform file system writes. The script allows specifying the output path via parameters, which could potentially be used to write data to unintended locations if the agent does not strictly follow the default path convention.
Audit Metadata