The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external neural recording data and specifically instructs the agent to perform visual analysis and curation based on generated plots. This creates a potential surface for indirect injection if data is maliciously crafted to produce visual patterns or text that the agent might interpret as instructions.
Ingestion points: The skill ingests data from external files such as .ap.bin, .meta, .oebin, and .nwb via si.read_spikeglx() and si.read_openephys() as shown in SKILL.md and scripts/neuropixels_pipeline.py.
Boundary markers: There are no explicit boundary markers or instructions to the agent to disregard potential instructions embedded within the processed data or visual reports.
Capability inventory: The skill has the capability to write files (e.g., si.export_to_phy in scripts/export_to_phy.py), execute external spike sorting binaries (e.g., si.run_sorter in scripts/run_sorting.py), and interact with external AI APIs.
Sanitization: Data is loaded using standard scientific libraries (spikeinterface, neo), but no specific security-focused sanitization of the recording content is performed.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to install various domain-specific scientific packages and clones repositories from GitHub.
SKILL.md and AUTOMATED_CURATION.md suggest installing packages such as spikeinterface, kilosort, bombcell, and unitmatch via pip or git clone from GitHub.
These references target established research tools and repositories on whitelisted domains.
[COMMAND_EXECUTION]: The skill facilitates the execution of external spike sorting software.
Scripts like scripts/run_sorting.py and scripts/neuropixels_pipeline.py invoke si.run_sorter(), which triggers the execution of external binaries (e.g., Kilosort4, SpykingCircus2) to process recording data.

neuropixels-analysis