The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): High-risk indirect prompt injection surface. The skill triggers on and processes untrusted user-provided database queries and code snippets while possessing powerful write and execution capabilities. 1. Ingestion points: User-supplied SQL, NoSQL, and C# code snippets (via SKILL.md trigger keywords). 2. Boundary markers: None present; the skill lacks instructions to delimit user input or ignore instructions embedded within queries. 3. Capability inventory: Bash, Write, Edit, and Task tools are enabled for optimization and analysis. 4. Sanitization: Absent; no instructions are provided to sanitize or validate query strings before they are processed or used in analysis tools.
[COMMAND_EXECUTION] (LOW): The skill enables the 'Bash' tool to facilitate database performance analysis, such as running explain plans or checking logs. While contextually relevant, this tool availability increases the potential impact of a successful injection attack.
[REMOTE_CODE_EXECUTION] (SAFE): No remote script downloads or dynamic execution of untrusted external code were identified.
[DATA_EXFILTRATION] (SAFE): No sensitive file access patterns or unauthorized network communication protocols were detected.
[OBFUSCATION] (SAFE): No encoded content, zero-width characters, or homoglyphs were found.

database-optimization