The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/sync_feishu_wiki.py executes the lark-cli (or a configured alternative) using subprocess.run. While the implementation correctly uses argument lists to prevent shell injection, the binary path itself is configurable via the YAML settings (provider.cli). This capability allows the execution of a local binary, which requires trust in the configuration and the provided CLI tool.
[EXTERNAL_DOWNLOADS]: The skill performs extensive network operations to fetch document content, metadata, and binary assets (images and attachments) from Feishu/Larksuite APIs and content delivery networks. Downloads are conducted using both the lark-cli tool and the Python urllib.request.urlopen library.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests large volumes of external data from the Feishu knowledge base and converts it into local Markdown files. If these documents contain malicious instructions, an agent reading the mirrored local content could be manipulated. The skill partially mitigates this by normalizing content and escaping non-standard HTML tags.
[DATA_EXFILTRATION]: No evidence of unauthorized data exfiltration was detected. The script's primary direction of data flow is from the cloud to the local environment. It includes a redact_output function designed to prevent local file paths and credentials from being written to logs or displayed in terminal output.
[SAFE_PRACTICE]: The script implements a file-based lock mechanism (sync_feishu_wiki.py:148) to prevent concurrent processes from corrupting the manifest and state files in the same output directory.

soia-cwork-feishu-doc-git-sync