The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses uv run --with to dynamically download and execute Python packages such as yake and scikit-learn from external registries at runtime.
[COMMAND_EXECUTION]: Multiple Bash heredoc blocks are used to execute shell commands including rg, fd, awk, and bc for file discovery and data processing.
[PROMPT_INJECTION]: The skill is vulnerable to script injection in Phase 5, where the $INPUT_FILE variable is interpolated directly into a Python script using single quotes. A malicious file path containing a single quote followed by Python commands could lead to arbitrary code execution.
[PROMPT_INJECTION]: There is a surface for indirect prompt injection (Category 8). Evidence: 1. Ingestion point: Reads terminal recording files (.txt) in Phase 5 and Phase 6. 2. Boundary markers: None are present to distinguish recorded text from agent instructions. 3. Capability inventory: Uses Bash and Python execution. 4. Sanitization: No sanitization or validation of the recording content is performed before analysis.
[DATA_EXFILTRATION]: The skill accesses and reads terminal recording files from the user's filesystem. While required for the skill's purpose, this represents a data exposure surface for local information.

asciinema-analyzer