The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes patterns for ingesting and processing external data from potentially untrusted sources (e.g., local CSV files, external API responses, and shared caches like Redis). While no malicious payload is present, the architectural surface allows for indirect prompt injection if an attacker-controlled file or API response contains instructions intended to influence the agent's behavior while it utilizes the skill's broad capabilities.
Ingestion points: The generate_large_file function reads from large_file.csv, CacheOptimizer retrieves data from Redis, and the frontend examples involve fetching results from an external fetchSearchResults API.
Boundary markers: The instructions and code snippets lack delimiters or instructions for the agent to ignore potential commands embedded within the processed data.
Capability inventory: The skill is configured with access to Bash (specifically for Python and pip operations), Read, Write, and Edit tools, allowing it to execute scripts and modify the filesystem based on analyzed data.
Sanitization: The provided Python and JavaScript snippets do not implement input validation, sanitization, or schema verification for the data being processed.

performance-expert