The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest user requirements and transform them into functional JSON pipeline configurations for MaaFramework. This creates an attack surface for indirect prompt injection where malicious instructions embedded in a user's request could be incorporated into the generated code. However, the skill provides extensive documentation on validation, observed evidence requirements, and structural hygiene that serves as a mitigation against accidental obedience to untrusted data.
Ingestion points: User-provided automation requirements and descriptions of game states.
Boundary markers: The skill emphasizes a 'hard rail' regarding 'observed vs guessed' states, though it lacks explicit sanitization instructions for user-provided strings within the JSON output.
Capability inventory: The skill facilitates the creation of configuration files that are subsequently executed by the MaaFW engine, which supports file operations, network calls, and custom code execution (Python, C++, Node.js, Go).
Sanitization: Not explicitly defined for runtime data interpolation, although schema validation is encouraged.
[COMMAND_EXECUTION]: The skill provides an instruction to install an external tool using npx skills add https://github.com/Windsland52/maafw-live --skill maafw-live --global. This involves the execution of a command to download and install a global skill. This operation is considered safe as the target repository is owned by the skill's author ('windsland52'), representing standard vendor-provided functionality for the intended development workflow.

maafw-pipeline