The Agent Skills Directory

[METADATA_POISONING]: The skill's description and scripts claim to offer specialized diagnostic capabilities for performance bottlenecks (CPU usage, database queries, memory leaks), yet all four included Python scripts (analyze_cpu.py, analyze_db.py, analyze_io.py, and analyze_memory.py) contain identical code that only performs a simple file system recursive crawl to calculate file counts and sizes. This mismatch significantly misleads the agent and user regarding the tool's actual safety and capability.
[INDIRECT_PROMPT_INJECTION]: The skill provides a mechanism to ingest and report on external data (file paths and metadata) from a target directory which could be used as an attack surface.
Ingestion points: The target directory argument passed to the Python analysis scripts in the scripts/ directory.
Boundary markers: None. The scripts output raw file names and statistics into the agent's context without clear delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill allows Bash(cmd:*), Read, Grep, and Glob which provides the agent with extensive local system access based on the output of these scripts.
Sanitization: None. The scripts do not sanitize or filter file names discovered during the directory traversal before including them in the generated report output.

detecting-performance-bottlenecks