The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_with_login_zsh.py executes commands within a login shell (zsh -lic) to facilitate loading environment variables like API tokens. It employs shlex.quote to sanitize inputs and prevent shell injection.
[COMMAND_EXECUTION]: The scripts/ankiconnect_import.py script executes external audio generation tools via the --audio-command parameter. It uses shlex.split on the command template and executes via subprocess.run with shell=False, which effectively mitigates command injection risks.
[PROMPT_INJECTION]: The references/word-coach-json-prompt.md file contains detailed instructions for a sub-agent to generate JSON data. While these are instructions for an LLM, they are intended for a controlled sub-agent environment to ensure data quality and formatting, rather than attempting to bypass the primary agent's safety filters.
[INDIRECT_PROMPT_INJECTION]: The skill processes data exported from the Eudic dictionary service. This data is subsequently processed by an LLM to generate Anki notes. While this represents a potential surface for indirect prompt injection if the user's dictionary data were to contain malicious instructions, the skill includes a validator (scripts/validate_trvs_coach_json.py) that checks for data integrity and specific content quality markers.

eudic-to-anki