The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands for environment exploration and task automation. Evidence includes calls to ls, grep, and sed on local project paths, execution of scripts/dt_helper.sh, and the creation and execution of temporary Bash scripts from /tmp for various API operations.- [DYNAMIC_EXECUTION]: The core workflow involves the agent dynamically generating Python test scripts (e.g., test_<module>_sdk.py and test_<module>.py) and executing them using uv run pytest. This involves runtime assembly and execution of code based on generated logic.- [DATA_EXFILTRATION]: The skill instructions expose a hardcoded absolute file path (/home/breath/project/personal/dingtalk-skills/tests/...), revealing internal directory structures and the system username 'breath'. Additionally, the skill's logic includes reading and parsing .env files, which are sensitive ingestion points for credentials and tokens.- [INDIRECT_PROMPT_INJECTION]: The skill processes external data such as SDK source files (client.py, models.py) and API error responses to extract technical specifications and required permissions. A compromised SDK or a malicious API endpoint could inject instructions into the parsing logic to influence the agent's behavior during skill creation.

dingtalk-skill-creator