The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_&_REMOTE_CODE_EXECUTION]: The skill instructs users to install the antom CLI by piping shell scripts from the official vendor domain mdn.alipayobjects.com (Alipay/Ant International) directly into bash or PowerShell's iex. These scripts are used to distribute the vendor's official command-line tooling.
[EXTERNAL_DOWNLOADS]: The skill dynamically fetches business knowledge, constraints, and capability definitions from cdn.marmot-cloud.com, a vendor-managed CDN. It also downloads settlement report files from dashboard.antom.com for local processing.
[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/retrieval/base_api_client.py and other modules to execute the antom CLI tool for authenticated report queries and transaction lookups.
[DYNAMIC_EXECUTION]: The scripts/cli.py launcher employs __import__ to dynamically verify that necessary Python dependencies like openpyxl and requests are installed before proceeding with data analysis tasks.
[INDIRECT_PROMPT_INJECTION]: The skill processes financial reports (CSV and XLSX) that may originate from external sources.
Ingestion points: The parse_reports function in scripts/core/parser.py ingests data from local and downloaded files.
Boundary markers: The parsing logic does not implement specific delimiters to separate potentially untrusted report content from agent instructions.
Capability inventory: The skill possesses capabilities for network downloads (requests) and local command execution (via the antom CLI).
Sanitization: Report data is parsed into structured formats using Decimal for precision; however, no sanitization specifically designed to mitigate prompt injection within report content is present.

antom-reconciliation-expert