The Agent Skills Directory

[DYNAMIC_EXECUTION]: The _load_engine function in scripts/convert.py dynamically resolves the location of the toolkit's scripts directory using Path(__file__).resolve() and injects these paths into sys.path to import the dataclean and ingest modules.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data files (Excel, CSV, JSON, PDF) which are then interpreted by the agent and used to generate reusable "conversion cards."
Ingestion points: The convert_file function in scripts/convert.py utilizes ingest.read_any() to process source files provided by the user, incorporating this data into the agent's context.
Boundary markers: While the skill documentation emphasizes "sense-checking" and manual review, the engine does not implement technical boundary markers or isolation for untrusted data processed within the prompt.
Capability inventory: The skill possesses the ability to read arbitrary files via lookup tables and write files to the local file system through write_output and write_template.
Sanitization: The _validate function checks output data against format rules (IBAN, regex, length), but does not perform sanitization to detect or neutralize natural language instructions embedded within the data.

data-convert