Remembering Conversations
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [PERSISTENCE]: The skill includes an
install-hookscript that modifies the local environment by adding an entry to~/.claude/hooks/sessionEnd. This allows the indexer to run automatically after each session ends, which is the intended mechanism for the 'auto-indexing' feature. - [INDIRECT_PROMPT_INJECTION]: The skill indexes and processes historical conversation data, which is an untrusted input surface. Malicious instructions in past conversations could potentially influence the summarizer or search subagent. The skill implements mitigations by using hierarchical summarization and specific subagent instructions (
tool/prompts/search-agent.md) that mandate synthesis over raw excerpts to reduce the risk of instruction leakage. - [DATA_EXPOSURE]: The skill reads local conversation logs and sends text segments to the Anthropic API via the
@anthropic-ai/claude-agent-sdkto generate summaries. This data flow is documented and necessary for the AI-assisted search and summarization features.
Audit Metadata