The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data which could contain malicious instructions to influence agent behavior.
Ingestion points: As described in SKILL.md, the agent reads project specifications, tracker context, project documentation, and comment threads.
Boundary markers: The instructions do not define clear boundaries or 'ignore' directives to prevent the agent from obeying instructions embedded within these external sources.
Capability inventory: The skill utilizes tools such as write_html, create_tokens, set_tokens, update_styles, and rename_nodes to modify design files.
Sanitization: No explicit content sanitization or validation of the ingested external data is mentioned.
[EXTERNAL_DOWNLOADS]: The skill configuration points to a local network endpoint for its core functionality.
Source: The agents/openai.yaml file specifies a connection to http://127.0.0.1:29979/mcp.
Execution method: The skill uses a streamable_http transport to interact with the local MCP server.
Context: This represents a standard interface for local tool execution.

paste-up