The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external configuration files that could potentially contain malicious instructions.
Ingestion points: The skill reads YAML, JSON, TOML, and YAML files from the configs/, recipes/, and experiments/ directories using cat and diff commands.
Capability inventory: The agent has access to file system tools (ls, cat, find, diff, rm), network monitoring trackers (wandb, neptune, mlflow), and job scheduler commands (scancel).
Boundary markers: The skill does not implement specific delimiters or 'ignore' instructions when reading the contents of these configuration files.
Sanitization: No sanitization or validation of the content within the configuration files is mentioned before the data is presented to the agent context.
[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands to verify the environment and manage artifacts.
Evidence: Commands such as find, diff, ls, and cat are used to inspect the local filesystem and configuration state.
Evidence: The skill provides instructions for the recursive deletion of directories using rm -rf for both local and remote (NFS/object store) paths during the cleanup phase.
Evidence: The skill interacts with the SLURM workload manager using the scancel command to terminate job reservations.

launch