The Agent Skills Directory

[COMMAND_EXECUTION]: Uses uv run to initiate training entrypoints, launch monitoring dashboards, and run checkpoint conversion scripts. It also employs shell utilities like grep, tail, wc, cat, ps, zstd, and jq for log analysis and data processing.
[COMMAND_EXECUTION]: Directs the agent to use srun and ssh for inspecting training and inference processes on remote nodes within a SLURM cluster.
[EXTERNAL_DOWNLOADS]: Employs uv sync to install and update project dependencies and environment-specific packages from the primeintellect-ai vendor and public registries.
[INDIRECT_PROMPT_INJECTION]: The skill involves reading and analyzing output from training logs and environment workers, which are sources of untrusted data.
Ingestion points: Log files located in {run_dir}/logs/latest/ and trace files in {run_dir}/monitors/file/.
Boundary markers: None; the agent is instructed to filter logs using grep or regular expressions for warning and error patterns.
Capability inventory: Broad command execution surface including uv run, srun, ssh, and network operations via curl for local health checks.
Sanitization: No evidence of input validation or sanitization for log content before it is processed by the agent.

training