read-claude-web-conversation
Pass
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local system commands using Python's
subprocessmodule and macOSosascript. These commands are used to identify running Chrome instances (viaps) and to inject JavaScript into the browser (via AppleScript). These operations are limited to facilitating the export process and do not include attempts at privilege escalation or persistence. - [SAFE]: The skill processes untrusted conversation data from the web. To mitigate the risk of indirect prompt injection or data corruption, it employs a 'fidelity gate' in
scripts/render_transcript.pythat audits the rendered output against the raw API payload. It also uses structural markdown delimiters and fencing to isolate the exported content from the agent's core instructions. - [SAFE]: The skill requires a browser security toggle ('Allow JavaScript from Apple Events') to be manually enabled by the user for certain fallback features. The documentation explicitly advises against automating this toggle, respecting standard security boundaries.
Audit Metadata