conversation-archive
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes platform-specific
gbrainCLI tools to ingest transcripts, synchronize files, and extract structured facts. Specifically, it instructs the agent to usegbrain transcripts ingest,gbrain sync,gbrain import, andgbrain extract-conversation-facts. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external chat exports and agent session logs, which are untrusted data sources that could contain embedded instructions intended to manipulate the agent.
- Ingestion points: Third-party export files (e.g.,
conversations.jsonfrom ChatGPT or Claude) and session logs are processed from the filesystem (SKILL.md). - Boundary markers: The skill specifies using
**Speaker:**formatting to separate dialogue and explicitly instructs the agent to treat imported content as data only, flagging any imperatives detected during conversion. - Capability inventory: The skill possesses the ability to write files to the
conversations/directory and execute local CLI commands for data enrichment. - Sanitization: A mandatory secret and PII redaction process is defined, using regex patterns and human review to replace sensitive strings with labeled placeholders before indexing.
Audit Metadata