The Agent Skills Directory

[SAFE]: The skill defines a legitimate and well-structured financial auditing workflow. It explicitly mandates a 'Safety Isolation Principle,' which ensures that external data sources like custodian reports are used solely for comparison and reporting, prohibiting direct entry or automated adjustments to the ledger without independent review.\n- [COMMAND_EXECUTION]: The skill uses read-only SQL queries to aggregate balances from core accounting systems (Step 1). These queries are standard for financial reporting and are parameterized by time period and account type to ensure scope control.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it processes external data from custodian reports (CSV/PDF) and financial data APIs (Tushare, SEC EDGAR).\n
Ingestion points: Data enters the agent's context through custodian reports processed via OCR tools and financial data retrieved via MCP interfaces.\n
Boundary markers: The workflow implements a 'Critic Review' phase (Step 4) that requires cross-verification of every variance with at least two data sources, serving as a manual security boundary.\n
Capability inventory: The skill is capable of executing database queries and drafting financial exception reports based on the processed data.\n
Sanitization: The skill relies on structured reporting templates and human-in-the-loop sign-off processes (Step 5) rather than automated technical sanitization of external inputs.

gl-reconciler