session-store-history
Pass
Audited by Gen Agent Trust Hub on Nov 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process historical data from the
session_storedatabase, including past user messages, assistant responses, and checkpoint summaries. This data is untrusted as it may contain malicious content from previous external sources that could attempt to influence the agent's current instructions during synthesis. - Ingestion points: The skill queries multiple tables in the
session_storedatabase, specifically theturns(user/assistant messages) andcheckpoints(summaries) tables. - Boundary markers: The instructions do not define specific delimiters or escaping mechanisms to separate retrieved historical content from the agent's current context.
- Capability inventory: The agent has access to a read-only SQL tool to perform the history search.
- Sanitization: No specific sanitization logic is prescribed for the content retrieved from the database beyond the instruction to synthesize findings into a short answer.
- [COMMAND_EXECUTION]: The skill's validation section includes instructions to execute local Python scripts using the shell.
- Evidence: The
SKILL.mdfile contains instructions to runpython _shared/run-trigger-evals.pyandpython _shared/run-functional-evals.pyfor testing purposes. - Context: These commands appear intended for development-time validation of the skill's logic rather than runtime behavior for the end-user, but they represent an execution surface for local code.
Audit Metadata