The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze external data files, which creates a surface for indirect prompt injection if those files contain malicious instructions.
Ingestion points: The skill reads and inspects rollouts.jsonl, *_materialized_inputs.jsonl, and *_reward_profiling.jsonl as described in SKILL.md and references/output-format.md.
Boundary markers: There are no explicit instructions or delimiters defined to isolate the content of these data files or to warn the agent against executing instructions contained within them.
Capability inventory: The skill utilizes command-line tools (gym env start, gym eval run, gym eval profile) and executes shell scripts as shown in references/quick-start.md.
Sanitization: No sanitization or validation logic is mentioned for the content of the processed JSONL files.

nemo-gym-reward-profiling