The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input describing plans and builds during the GRILL phase. It possesses capabilities to explore the codebase (read) and write plan files to disk. There are no explicit boundary markers or sanitization steps defined in the instructions to mitigate the risk of instructions embedded within user data influencing the agent's behavior.
Ingestion points: User-provided project descriptions and answers to interview questions in the GRILL phase.
Boundary markers: Absent; the instructions do not specify delimiters or warnings to ignore instructions within user plans.
Capability inventory: Filesystem read access ("explore the codebase"), web search capabilities, and filesystem write access ("write it to a location they name").
Sanitization: Absent; no mention of escaping or filtering external content.
[EXTERNAL_DOWNLOADS]: The README.md file contains installation instructions using npx to fetch the skill from the author's GitHub repository (m1nga/skills). This is a standard installation procedure provided by the vendor.
[COMMAND_EXECUTION]: The skill utilizes tool capabilities to read from the local codebase to answer design questions and writes output files to the local filesystem to save generated project plans. These operations are aligned with the skill's primary purpose of project scoping and documentation.

grilling