The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell command templates for directory management (mkdir), file duplication (cp), and file removal (rm). These commands use placeholders like {experiment-name} which are intended to be replaced by the agent during the workflow.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill instructs the agent to copy source code into an experiments folder. While it includes a step to automate the addition of this folder to .gitignore, failure to execute this correctly could lead to accidental exposure of sensitive local code if the directory is inadvertently committed to a repository.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: User-provided experiment names and hypothesis metadata in SKILL.md.
Boundary markers: None identified in the provided templates.
Capability inventory: Shell execution capabilities for file and directory manipulation across SKILL.md.
Sanitization: No explicit validation or sanitization is requested for the user-supplied strings before they are used in shell commands.
[SAFE]: The skill contains no obfuscation, remote execution patterns, or hardcoded secrets. Its functionality is consistent with its described role as a developer utility for technical exploration.

tinkering