The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes reconciliation data ("break row") which could contain malicious instructions designed to influence the agent's logic during the tracing process.
Ingestion points: External data enters the context via the "break row" input described in SKILL.md.
Boundary markers: There are no explicit delimiters or boundary markers (such as XML tags or unique markers) defined to isolate the untrusted input data from the skill's core instructions.
Capability inventory: The skill utilizes internal-gl and subledger MCP tools to perform lookups on internal financial systems based on the provided input data.
Sanitization: The instructions do not include validation or sanitization logic to filter potentially malicious strings from the input before processing.

break-trace