The Agent Skills Directory

[SAFE]: The skill defines a standard organizational workflow for experiment tracking without the use of external dependencies or remote code.
[DATA_EXFILTRATION]: All data operations are local to the project filesystem; there are no network calls or external data transfers.
[COMMAND_EXECUTION]: The skill does not execute shell commands or use any system-level tools.
[PROMPT_INJECTION]: The skill reads from local files experiments/journal.md and experiments/lessons.md, creating a potential surface for indirect prompt injection.
Ingestion points: local markdown files.
Boundary markers: None.
Capability inventory: File read/write.
Sanitization: None. However, since the data is intended to be generated and managed by the agent itself for logging purposes, this is assessed as safe practice within this context.

ml-experiment