The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes command-line operations using the opensquilla CLI tool (e.g., opensquilla swebench solve, opensquilla swebench pull). These commands are part of the vendor's own software suite and are used to manage the benchmark lifecycle.
[EXTERNAL_DOWNLOADS]: The skill installs the opensquilla[swebench] package via uv or pip. As this package belongs to the skill's author (OpenSquilla), it is categorized as a standard vendor resource. Additionally, the skill pulls official benchmark images from Docker Hub, which is expected behavior for its primary function.
[CREDENTIALS_UNSAFE]: The skill requires the OPENROUTER_API_KEY environment variable. This key is used to allow the agent to interact with language model providers via OpenRouter. The management of this key is handled by the platform configuration in the skill's frontmatter.
[DATA_EXFILTRATION]: While the skill accesses external datasets (e.g., from HuggingFace) and interacts with the OpenRouter API, no patterns indicating unauthorized data harvesting or exfiltration were found. The network activity is limited to downloading benchmark data and communicating with LLM endpoints.

swe-bench