The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes builder input during an "interview" phase to generate task markdown files. These files are saved to the repository's tasks/ directory and published to a team catalog via the save_task tool. This pattern creates a surface for indirect prompt injection where malicious input could potentially influence the behavior of the generated tasks or downstream evaluation agents.
- Ingestion points: SKILL.md (Step 2: Interview the builder).
- Boundary markers: The skill uses markdown headers (## Prompt, ## Success Criteria) to structure the output, but there is no explicit instruction to sanitize or escape the user-provided descriptions of agent behavior or "fears."
- Capability inventory: File system write access to the manifest's tasks/ directory and network communication via the Pome control MCP server tools (save_task).
- Sanitization: The instructions do not specify any validation or filtering of user input before it is interpolated into the generated markdown files.
[EXTERNAL_DOWNLOADS]: The documentation references an installation command using npx to fetch the pome-sh/digital-twins skill set and establishes a connection to a Pome control MCP server at https://mcp.pome.sh/mcp. These resources are part of the vendor's provided infrastructure for the skill's operation.

pome-author-task