The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including source code (guidelines/from-code.md), functional requirements (guidelines/from-requirements.md), and TLA+ specifications (guidelines/from-tlaplus.md). While this creates an attack surface for indirect prompt injection, the agent uses this data to generate formal specifications rather than executing it directly. Ingestion points include user-provided source files and documentation. The guidelines do not explicitly require the use of delimiters or warnings when processing external content. Capability inventory includes the use of the quint CLI tool to evaluate the generated output. The skill focuses on abstraction and extraction, which naturally filters implementation details.
[COMMAND_EXECUTION]: The skill correctly uses the quint command-line utility to verify the correctness of models. The commands used (quint typecheck, quint run, quint test) are standard operations for the Quint modeling language and are used as intended within the development workflow. These commands are executed locally on generated files to ensure their formal correctness.

quint-modeling