The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/run.sh script (detected as a downloader by AV scanners) automatically identifies the host platform, downloads a corresponding binary from the geekjourneyx/md2wechat-skill GitHub repository, and executes it. This provisioning mechanism lacks integrity verification such as checksum validation, creating a supply-chain risk.\n- [REMOTE_CODE_EXECUTION]: Installation instructions in QUICKSTART.md and the scripts/install.sh utility use the high-risk curl | bash pattern. On Windows, scripts/install.ps1 employs Invoke-Expression (iex) to execute a remote script. Both patterns are high-risk as they execute unverified remote code directly in the shell.\n- [COMMAND_EXECUTION]: Several scripts, including build.py and generate_image.py, utilize the subprocess module to execute shell commands. This includes running the dynamically downloaded binary and managing the playwright browser driver.\n- [CREDENTIALS_UNSAFE]: The setup.py configuration wizard and project documentation (docs/CONFIG.md) guide users to store sensitive WeChat AppID and AppSecret credentials in a plaintext YAML file (md2wechat.yaml), which increases the risk of credential exposure.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted Markdown content (via build.py and convert.go) and interpolates it into prompts for LLM-based layout generation. While it uses boundary markers and implements basic filtering for tags like <script>, the combination of untrusted input and sensitive capabilities (network access, shell execution) presents a vulnerability surface.\n- [EXTERNAL_DOWNLOADS]: The skill performs extensive network operations, fetching binaries from GitHub, generating images via external AI services (ModelScope, OpenAI, TuZi), and interacting with WeChat's official API servers.

md2wechat