The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bash commands for linting, building, and testing across various languages (Go, Python, TypeScript). These operations are restricted to specific command prefixes, which minimizes the risk of arbitrary command execution.
[DATA_EXFILTRATION]: The skill analyzes command output and source code files to identify root causes of bugs. While this involves processing untrusted data (Indirect Prompt Injection surface), the skill lacks network exfiltration capabilities and is restricted by the platform's tool configuration.
Ingestion points: Log output from commands like make, pytest, and ruff, as well as source code file contents.
Boundary markers: No explicit delimiters are used when processing external data.
Capability inventory: Restricted bash execution, file system reading, and task management tools.
Sanitization: No explicit sanitization or filtering of external content is performed before processing.

fixing-code