The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill includes SQL scripts and instructions to access TiDB system tables (e.g., information_schema.slow_query, processlist, stats_meta). It also references the TiDB Debug API (port 10080) to dump statistics and schema metadata. These operations are strictly related to the skill's primary objective of database performance analysis and do not involve unauthorized exfiltration of sensitive user data.
[COMMAND_EXECUTION]: The script scripts/generate_tidb_issue_experiences.py executes the GitHub CLI (gh) via subprocess.run. This is used to fetch issue data from the official TiDB repository. The implementation uses argument lists rather than shell strings, minimizing injection risks, and serves as a legitimate tool for local data generation.
[INDIRECT_PROMPT_INJECTION]: The skill provides a pipeline for ingesting untrusted data from external sources.
Ingestion points: Content is fetched from GitHub issues using the generate_tidb_issue_experiences.py script.
Boundary markers: The fetched data is organized into structured markdown files (references/tidb-customer-planner-issues/).
Capability inventory: The skill is intended to help agents generate query tuning SQL and diagnostic commands.
Sanitization: The Python script includes a sanitize_text function that strips HTML comments, code blocks, and images from fetched issue bodies, which significantly reduces the risk of malicious instructions being processed by the agent.

tidb-query-tuning