The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches pre-compiled binaries and Docker images from an unverified GitHub repository (github.com/JulienTant/blogwatcher-cli) and the GitHub Container Registry.
[COMMAND_EXECUTION]: Instructs the user to install the tool by extracting binaries directly into /usr/local/bin, a privileged system path. It also performs filesystem operations like moving database files in the user's home directory (mv ~/.blogwatcher/blogwatcher.db ~/.blogwatcher-cli/blogwatcher-cli.db).
[PROMPT_INJECTION]: Indirect Prompt Injection Surface:
Ingestion points: The tool consumes and displays data from untrusted external sources, specifically RSS and Atom feeds, during the scan and articles commands (defined in SKILL.md).
Boundary markers: No boundary markers or instructions are provided to help the agent distinguish between tool output and instructions embedded within the feed content.
Capability inventory: The tool performs network operations and local filesystem writes (SQLite database), which could be manipulated if an agent follows instructions from a malicious feed.
Sanitization: There is no documented validation or sanitization of the feed content before it is presented to the agent.

blogwatcher