The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from various external financial sources, which could contain malicious instructions designed to influence the agent's behavior during the update process.
Ingestion points: External data enters the agent context via earnings transcripts, KPI releases, and regulatory filings (10-K, 10-Q, 8-K) during Step 1.
Boundary markers: No specific programmatic delimiters are defined for the ingested data; however, Step 3 mandates a human-in-the-loop review where the agent must present a mapping table to the user before any cell is written.
Capability inventory: The skill possesses capabilities for file system operations (cp), integrity verification (shasum), and execution of specialized internal Python scripts for workbook manipulation.
Sanitization: Relies on a manual verification step in Step 3 and Step 4, where the agent logs every write and assumption for user audit.
[COMMAND_EXECUTION]: The skill instructions invoke several shell commands and internal scripts to manage the financial model lifecycle.
Evidence: Step 2 utilizes shasum -a 256 for file integrity and cp for version control.
Evidence: Step 4 and Step 7 execute internal Python utilities (insert.py, recalc.py, audit.py) located within the relative .agents/skills/ directory to perform structural edits and audits.

model-update