conversation-archive
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted third-party data from chat exports (ChatGPT, Claude, Perplexity) and agent session transcripts. This content is converted into markdown pages and stored in the brain repository.
- Ingestion points: Chat history files such as
conversations.jsonand session logs. - Capability inventory: Writing markdown pages to
conversations/, performing automated fact extraction, and executing CLI tools for synchronization and import. - Boundary markers: The instructions explicitly state that transcript bodies are DATA and should not be interpreted as instructions. It mandates flagging agent-directed imperatives during the conversion process to prevent them from being executed as tasks.
- Sanitization: The skill enforces a mandatory pre-write redaction step for secrets (API keys, JWTs, private keys) and PII (emails, addresses) using deterministic patterns.
Audit Metadata