The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill maintains and ingests state from a project file and external tool outputs, which can act as a vector for malicious instructions.
Ingestion points: The skill reads the _operator/KONTINGENT.md ledger file at the start of new sessions to establish rules and history. It also processes data from /idea-mining and /skill-finder during the pre-escalation procedure.
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore potentially embedded instructions within the ingested data.
Capability inventory: The skill possesses file writing capabilities (creating and updating the ledger) and triggers subsequent model calls based on its analysis.
Sanitization: No sanitization or validation logic is defined for the content read from the ledger file or received from external tools.

budgetierung