The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates state transfer between chat sessions by ingesting user-provided text and outputting it for later ingestion by the agent. This creates a surface for instructions to be embedded in the data stream.
Ingestion points: The scripts/gen.py script reads handoff content from stdin or a file specified via the --from argument.
Boundary markers: The skill does not use delimiters or include instructions to the reading agent to ignore embedded commands within the handoff data.
Capability inventory: The agent utilizing this skill is intended to perform complex file and system operations based on the inherited task description.
Sanitization: The input content is structured into sections but is not sanitized to remove or neutralize natural language instructions.
[COMMAND_EXECUTION]: The skill requires the agent to execute local Python scripts and manage a temporary local web server.
Evidence: SKILL.md defines workflows involving python3 scripts/gen.py and python3 scripts/read.py.
Details: gen.py starts a local HTTP server on 127.0.0.1 to handle save requests from the generated HTML interface. The server includes a 5-minute idle timeout for safety.
[DYNAMIC_EXECUTION]: The skill generates executable HTML and JavaScript content at runtime based on user input.
Evidence: scripts/gen.py uses f-string interpolation to build an HTML document containing the handoff data and a JavaScript-based editor. The interpolation of the document title lacks proper HTML escaping, presenting a minor Cross-Site Scripting (XSS) surface in the generated file.

switch-chat