The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided JSON files (claims and evidence excerpts) which are then processed by the scripts/jev_office_gate.py script and included in an API payload. This creates an attack surface where maliciously crafted data could attempt to influence the agent's decision-making or manipulate the downstream API response.
Ingestion points: The scripts/jev_office_gate.py script reads data from a JSON file path provided by the user.
Boundary markers: The data is encapsulated in a structured JSON payload, but the text content itself lacks explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The script performs network POST requests to an external API endpoint.
Sanitization: Basic normalization is applied to identifiers, but the claim and evidence text are not sanitized or escaped to prevent injection patterns.
[DATA_EXFILTRATION]: The script scripts/jev_office_gate.py performs network operations to api.typesafe.ai to send citation data for review. This domain is not included in the standard whitelist of trusted services. The script transmits claim and evidence data to this external endpoint, which constitutes a standard operational flow for the skill but falls under the classification of network operations to non-whitelisted domains.

jev-office-gate