The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads and installs the aiperf benchmarking harness from a third-party GitHub repository.
Evidence: python -m pip install 'aiperf @ git+https://github.com/SemiAnalysisAI/agentx-harness.git@56a0cf70f4c0359454ee4bd15a17770b541a3e3e'
The installation uses a specific commit hash (SHA) to ensure the integrity and consistency of the external code, which is a security best practice for managing external dependencies.
[COMMAND_EXECUTION]: The skill executes shell commands to configure the environment and run the profiling tool.
Evidence: Uses aiperf profile to initiate benchmarking scenarios based on user-defined environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill processes external datasets that may contain untrusted data.
Ingestion points: The aiperf tool downloads and caches public datasets and traces as defined by the AGENTX_DATASET variable in SKILL.md.
Boundary markers: The instructions advise the agent never to silently truncate requests, but do not specify specific sanitization for dataset content.
Capability inventory: The skill utilizes network access to communicate with inference endpoints and shell execution to run the aiperf suite (documented in SKILL.md).
Sanitization: The skill includes an explicit security check, advising against the use of --tokenizer-trust-remote-code unless the user has manually verified and trusted the repository.

run-agentx