The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from database logs and system introspection tables, which serves as a vector for indirect prompt injection. A malicious actor who can write to database logs or influence system table content (e.g., through query comments) could embed instructions aimed at redirecting the agent's behavior during the diagnostic process.
Ingestion points: The skill instructs the agent to capture exception strings from client returns, use Grep to extract lines from log files, and query system.* tables (like system.query_log and system.processes).
Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" warnings when processing the captured log data or query results.
Capability inventory: The skill utilizes Bash(curl:*) for interacting with the ClickHouse HTTP interface and provides various SQL commands for execution, including ALTER TABLE, SET configuration, and KILL QUERY operations.
Sanitization: There is no evidence of content sanitization, validation, or escaping of the external data before it is processed by the agent.

clickhouse-common-errors