The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions suggest installing companion tools (grilling, research) from an external GitHub repository (github.com/mattpocock/skills) using the npx command. This source is not recognized as a primary trusted organization or a well-known enterprise cloud service provider.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability due to how it handles external data during the planning process.
Ingestion points: The skill reads untrusted data including 'user feedback', 'incidents', 'research', and 'backlog' items from the repository to generate its outputs.
Boundary markers: There are no instructions provided to the agent to use delimiters or ignore potentially malicious instructions that might be embedded within the external data sources.
Capability inventory: The agent is instructed to write and update Markdown artifacts and generate task specifications that define future implementation work based on this data.
Sanitization: The instructions do not include any steps for the agent to sanitize, escape, or validate the ingested content before it is incorporated into the planning artifacts.

roadmap-planning