The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from user-provided state and questions, which are then transmitted to an LLM-based API for evaluation.
Ingestion points: The scripts/create_decision.sh script accepts data via --state, --state-file, --questions, and --questions-file arguments.
Boundary markers: The implementation does not explicitly define prompt boundary markers or instructions to disregard embedded commands in the user-supplied state.
Capability inventory: The skill utilizes curl for network requests to the Frevana API and supports file writing through the --output parameter.
Sanitization: The script performs structural validation of JSON and enforces specific question schemas (choice, noul, score) using an internal Python block.
[DYNAMIC_EXECUTION]: The bash script create_decision.sh contains an embedded Python script used for JSON normalization and request validation. The Python code is static and defined within the shell script to avoid external dependencies like jq.
[COMMAND_EXECUTION]: The skill executes local shell commands via create_decision.sh to perform its primary tasks. Additionally, the test suite in tests/test_create_decision.py uses subprocess.run to execute the shell script during integration testing.

jev-latest