The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes content from arbitrary external URLs (via r.jina.ai, defuddle.md, WeChat, and Feishu) and provides it to the agent. This creates an attack surface where a malicious website could include hidden instructions designed to influence the agent's behavior during summarization or subsequent tasks.
Ingestion points: External content fetched via curl, scripts/fetch_weixin.py, and scripts/fetch_feishu.py.
Boundary markers: The skill instructions do not specify the use of delimiters or warnings to treat the fetched content as untrusted data.
Capability inventory: The agent has the ability to write files to ~/Downloads/, execute local Python scripts, run shell commands (curl, npx), and modify the user's Obsidian vault via obsidian-cli.
Sanitization: No explicit sanitization or filtering of the fetched Markdown content is defined in the skill logic.
[COMMAND_EXECUTION]: The skill makes extensive use of local command execution to perform its tasks.
It executes curl to interact with the Jina AI and Defuddle proxy services.
It invokes python3 to run internal scraping scripts for WeChat and Feishu.
It uses npx agent-fetch as a fallback mechanism for content extraction.
It executes obsidian-cli to create and update notes in the user's Obsidian vault, including an instruction to perform this action by default even if not explicitly requested by the user.
[EXTERNAL_DOWNLOADS]: The skill retrieves data from several external services.
Fetches content from https://r.jina.ai and https://defuddle.md.
Interacts with https://open.feishu.cn for document retrieval.
The README.md directs users to install external dependencies like Playwright and its associated browser binaries.

markdown-proxy