The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill is designed to execute shell commands using Bun Shell.
Evidence: scripts/main.ts uses the $ template literal to run git log and grep with user-provided arguments.
Context: The implementation uses Bun's auto-escaping feature (e.g., ${query}), which mitigates traditional shell injection where special characters might break out of the command string.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill accepts external arguments ($ARGUMENTS) and processes them via grep and git.
Ingestion Points: process.argv.slice(2) in scripts/main.ts ingests data directly from the agent's prompt interpolation.
Boundary Markers: None present in the current template; input is passed directly to the script.
Capability Inventory: Subprocess execution via Bun Shell ($).
Sanitization: Relies on Bun's internal template literal escaping. While this prevents shell-level escaping, it does not prevent the LLM from being influenced by malicious content returned from git log or grep results (Indirect Injection).

template