The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a configuration manager that translates untrusted user requirements (e.g., supervisor preferences) into persistent instructions within the workspace guidelines.md file. These instructions are intended to be honored by other skills in the proposal workflow, creating a surface for indirect prompt injection.
Ingestion Points: User-supplied requirements and instructions provided via chat interactions.
Boundary Markers: The skill enforces a structured TOML block within guidelines.md for mechanical settings and uses separate freeform prose for general guidance.
Capability Inventory: The skill is authorized to modify the guidelines.md file in the workspace root, which influences the execution logic of the entire proposal skill suite.
Sanitization: The skill performs explicit validation on the [paths] proposals configuration to block directory traversal attempts (..), absolute paths, and home directory shortcuts (~).

proposal-customize