The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to automatically ingest tax and financial data from a file named foyer.json at the project root to establish user context. This data is then used in reasoning and passed as arguments to the calc_ir.py and dgfip_oracle.py scripts. As this file represents untrusted input that influences the agent's logic while the agent possesses script execution and network capabilities, it creates a surface for indirect prompt injection.\n
Ingestion points: Reads foyer.json and examples/*.json to build the tax profile context.\n
Boundary markers: The instructions do not specify strict delimiters or warnings to ignore instructions embedded within the user-provided JSON data.\n
Capability inventory: The skill uses scripts/calc_ir.py for local deterministic calculations and scripts/dgfip_oracle.py for network-based verification against official tax simulators.\n
Sanitization: There is no explicit sanitization or validation logic described in the instructions for the data ingested from foyer.json before it is used in prompts.

fiscaliste