The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (detect_system.py and benchmark_ollama.py) to profile system hardware, running commands such as nvidia-smi, sysctl, and wmic. It also generates shell commands to configure environment variables and manage the Ollama service.
[PRIVILEGE_ESCALATION]: The skill utilizes sudo on Linux to create systemd drop-in configuration files and restart system services. It implements safety guidelines requiring explicit user consent before executing these commands.
[PERSISTENCE]: To ensure optimizations survive reboots, the skill modifies shell initialization files like ~/.zshrc or ~/.bashrc and creates persistent systemd units. It mitigates the risk of system instability by creating backups of modified configuration files.
[DYNAMIC_EXECUTION]: The skill contains a 'Self-Improve' section that authorizes the agent to modify the skill's own instructions or reference documents based on observed environmental issues or user preferences.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from hardware profiling scripts and user-provided inputs. While the data is parsed as JSON, the resulting values directly influence the agent's logic for tier classification and system modification, presenting an attack surface for indirect influence over agent behavior.

ollama-optimizer