The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external geophysical data formats (CSV, SEG-Y, EDI), which creates a potential surface for indirect prompt injection.
Ingestion points: Data is loaded via pandas.read_csv in scripts/ert_layered_diagnostic.py and through domain-specific libraries like gprpy and mtpy as described in SKILL.md.
Boundary markers: The workflow relies on strict structured data formats and explicit schema definitions provided in references/validated_branches.md.
Capability inventory: The skill's capabilities are limited to local file read/write operations and scientific computations. It does not utilize network access tools or shell execution commands.
Sanitization: The load_inputs function in scripts/ert_layered_diagnostic.py implements rigorous validation, including numeric type enforcement, finiteness checks, and electrode identity verification, which mitigates the risk of processing malicious payloads embedded in data.
[SAFE]: The skill incorporates several proactive security and integrity measures:
Data Integrity: The Python script uses hashlib to generate and verify SHA256 checksums for input and output files, ensuring that the processed data remains untampered.
Network Privacy: The documentation in references/validated_branches.md specifically instructs the agent to set get_elevation=False when using the mtpy library to prevent unexpected external network calls for elevation metadata lookup.

near-surface-geophysics