The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill fetches and executes installation scripts (install.sh for Linux/macOS and install.ps1 for Windows) from the official ROCm GitHub repository. This occurs during the initialization phase if the rocm CLI tool is not already detected on the system.
[COMMAND_EXECUTION]: The skill frequently executes shell commands via the rocm CLI (examine, diagnose, fix) to inspect the host system and apply configuration changes.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential vulnerability surface for indirect prompt injection:
Ingestion points: The skill accepts user-provided error text ("symptom") in the SKILL.md workflow.
Boundary markers: There are no explicit boundary markers or warnings to the agent to ignore instructions embedded within the symptom text.
Capability inventory: The agent has the capability to execute shell commands with potentially elevated privileges via rocm fix (which supports sudo).
Sanitization: The user's symptom is interpolated directly into a shell command: rocm diagnose --symptom "<paste the exact error>" --json. This lacks sanitization, allowing for potential shell command injection if the input contains shell-active characters like backticks, semicolons, or command substitutions.
[EXTERNAL_DOWNLOADS]: Fetches installation scripts and configuration data from the ROCm/rocm-cli repository on GitHub.

rocm-doctor