The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses the Chrome DevTools Protocol (CDP) to inject and execute dynamic JavaScript within an active browser session.
Evidence: In scripts/enrich_via_wechat_web_api.py, the _evaluate function utilizes the websockets library to send Runtime.evaluate commands to a browser tab.
Risk: This allows the skill to execute arbitrary code in a context containing sensitive session cookies and tokens, which is necessary for accessing private WeChat APIs but poses a risk of context abuse.
[COMMAND_EXECUTION]: The skill executes system-level commands to retrieve stored credentials.
Evidence: In scripts/publish_existing_draft.py, the credentials_from_oneshot_keychain function invokes the macOS security find-generic-password command.
Risk: Accessing the system keychain is a sensitive operation that could be abused to retrieve unauthorized credentials if the scope is not strictly controlled.
[INDIRECT_PROMPT_INJECTION]: The skill processes external article data which serves as a potential vector for indirect prompt injection.
Ingestion points: Article content (Markdown or HTML) is ingested from local files during the preflight and publishing stages (e.g., in scripts/publish_via_gateway.py).
Boundary markers: While the skill validates the structural integrity of the article (such as checking for specific benchmark sections or image ratios), it does not implement explicit boundary markers or instructions to the LLM to ignore embedded commands within the processed text.
Capability inventory: The skill possesses high-privilege capabilities including network exfiltration via a gateway, shell command execution, and browser session automation.
Sanitization: Although the content is parsed using BeautifulSoup for structural validation, there is no evidence of sanitization specifically targeted at neutralizing natural language prompt injection patterns.
[EXTERNAL_DOWNLOADS]: The skill communicates with the vendor's API gateway for its core publishing functionality.
Evidence: The script scripts/publish_via_gateway.py routes WeChat Official Account API requests through https://api.lovstudio.ai/wechat/official-account using a managed API key.

lov-publish-wechat-article