The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process complex multi-modal robotics datasets (MCAP, HDF5, LeRobot, etc.). This creates a vulnerability surface where malicious content embedded within the dataset files could attempt to influence the agent's logic during the conversion, survey, or blueprint creation phases.
Ingestion points: External data files described in SKILL.md under Step 1 ('obtain sample data') and Step 2 ('understand your data').
Boundary markers: No explicit delimiters or 'ignore' instructions are mandated for the processing pipeline to isolate data values from the agent's instruction context.
Capability inventory: The agent is guided to write and execute Python scripts for conversion and registration, perform file system writes (.rrd and .rbl files), and use the rerun CLI.
Sanitization: The workflow relies on standard Rerun/Arrow serialization; there are no instructions for sanitizing or filtering the content of the source data to prevent potential injection via metadata or string fields.
[EXTERNAL_DOWNLOADS]: Workflow Step 1 in SKILL.md explicitly directs the creation of download scripts to fetch datasets from remote sources. While this is the intended purpose of the skill, it involves network operations to retrieve arbitrary external data.
[COMMAND_EXECUTION]: The skill utilizes the rerun CLI tool and instructs the agent to write and execute Python scripts for data registration and conversion, as seen in references/registering.md. These are standard vendor-provided tools (rerun-io).

rerun-dataset-conversion