The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a 'Run Record Template' in SKILL.md that is designed to be populated with external data (e.g., project names, commit SHAs, hyperparameter dictionaries, and rejection reasons).
Ingestion points: The [FILL] placeholders in the Run Record Template section of SKILL.md represent points where untrusted or external data is interpolated into the agent's context.
Boundary markers: The template lacks explicit delimiters or instructions (e.g., XML-like tags or specific escape sequences) to prevent the agent from misinterpreting instructions embedded within the filled values.
Capability inventory: The skill does not include any scripts or defined tool invocations, but the platform's default capabilities (file access, network operations) remain accessible to an agent utilizing these instructions.
Sanitization: There are no instructions or logic provided to validate, filter, or sanitize the data being placed into the [FILL] fields.
[NO_CODE]: The skill consists entirely of Markdown instructions and documentation. No Python scripts, JavaScript files, or other executable binaries are included in the skill package.

Experiment Tracking