The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script scripts/check_words.py to interface with the analysis API.
Evidence: SKILL.md contains instructions for the agent to execute python3 scripts/check_words.py "待检测的文案内容" to perform the analysis.
[DATA_EXFILTRATION]: The skill transmits user-provided text content to an external vendor-owned API for analysis.
Evidence: scripts/check_words.py performs a POST request to https://doubaoya.com/api/apis/tool/check-banned-words/call. This is the intended primary functionality for the service provided by author zizhanovo.
[CREDENTIALS_UNSAFE]: The skill manages an API key via environment variables and includes explicit instructions to prevent leaking it.
Evidence: SKILL.md and scripts/check_words.py use the DOUBAOYA_API_KEY environment variable. The documentation includes a 'Safety Rule' to ensure the key is never displayed in logs or agent output.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted user text that is processed and returned by an external API.
Ingestion points: User-supplied text is passed as a command-line argument to the detection script (identified in SKILL.md).
Boundary markers: No specific delimiters or XML-style tagging are mandated for the input text in the prompt workflow.
Capability inventory: The skill possesses network access (via urllib in scripts/check_words.py) and command execution (running the script itself).
Sanitization: No explicit sanitization or filtering of the user text or API response is performed before being processed by the LLM for rewriting.

wechat-banned-words