The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted royalty statements (CSV, XLSX, PDF) as its primary input.
Ingestion points: Seller files in the source/ directory are read by normalization and extraction scripts.
Boundary markers: The skill treats source files as immutable evidence and follows strict normalization rules defined in its reference documentation.
Capability inventory: The skill writes to local workspace directories and executes validation scripts via subprocess.
Sanitization: Normalization scripts perform arithmetic cleaning and structured parsing, which inherently limits the risk of data confusion.
[DYNAMIC_EXECUTION]: Local scripts dynamically load sibling modules for code reuse.
Evidence: scripts/auto-column-map.py uses importlib.util to load normalize-royalty-statement.py at runtime. This is a standard Python pattern for loading scripts with non-identifier names.
[COMMAND_EXECUTION]: A readiness wrapper script executes other local validation scripts to verify deal data.
Evidence: scripts/run-deal-checks.py uses subprocess.run to call hardcoded internal scripts such as validate-deal-workspace.py using the current Python interpreter to ensure artifact consistency.
[EXTERNAL_DOWNLOADS]: The skill fetches catalog performance data from a vendor-owned API.
Evidence: The skill interacts with the Recoup Research API at https://api.recoupable.dev/api to retrieve streaming play counts and artist metadata. These resources are provided by the skill's author for catalog valuation purposes.

recoup-catalog-review-deal