The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestration script scripts/source_search.py uses subprocess.run to launch local Python sub-processes, such as moegirl_api.py. This pattern is used to run multiple research agents in parallel to gather character data. The commands are constructed using a list of arguments, which prevents standard shell-based command injection.\n- [EXTERNAL_DOWNLOADS]: The component scripts/moegirl_api.py connects to the Moegirlpedia API (zh.moegirl.org.cn) using urllib to fetch character lore and transcripts. This is a well-known community service for anime and game data, used here to provide factual grounding for character distillation.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by retrieving untrusted text from external wikis and storing it in the references/research/ directory. This content is later analyzed by the AI to generate behavior instructions. Maliciously crafted wiki entries could potentially influence the agent's behavior during the distillation process, despite the presence of quality checks.

csp