The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through stored experiment metadata.
Ingestion points: Data is retrieved from the experiments.jsonl file via the list_experiments and compare_experiments functions in __init__.py.
Boundary markers: No delimiters or instructions are used to separate untrusted experiment data from the agent's context.
Capability inventory: The skill has the capability to append data to a local file (experiments.jsonl) within the designated experiments directory in __init__.py.
Sanitization: The skill does not perform sanitization or validation on user-supplied fields like notes, tags, or the arbitrary data provided via the **kwargs catch-all in the log_experiment function.

experiment-tracker