The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data which could potentially contain malicious instructions, though the risk is minimal given the intended use case.
Ingestion points: The skill ingests user-provided text and captions through the swanlab.log function, specifically using swanlab.Text and the caption parameter for media objects in SKILL.md and references/visualization.md.
Boundary markers: No specific delimiters or instructions (e.g., "ignore embedded instructions") are used to isolate logged text or captions from the agent's control logic.
Capability inventory: The skill possesses network communication capabilities through the swanlab library to transmit logs and uses shell commands for CLI operations such as swanlab login, swanlab watch, and swanlab sync in SKILL.md.
Sanitization: The provided examples do not include explicit sanitization or validation of the text content or captions before they are passed to the logging API.
[EXTERNAL_DOWNLOADS]: The skill manages legitimate software dependencies from standard registries.
Fetches and installs official packages including swanlab, pillow, and soundfile using pip.
References standard machine learning frameworks such as transformers, pytorch-lightning, and fastai for integration.
[COMMAND_EXECUTION]: The skill utilizes command-line tools for experiment management.
Employs the swanlab CLI for authentication (swanlab login), local dashboard monitoring (swanlab watch), and data synchronization (swanlab sync) in SKILL.md.

experiment-tracking-swanlab