The Agent Skills Directory

[PROMPT_INJECTION]: The skill generates instruction files that override default agent behavior using 'non-negotiable' principles and operational defaults like 'Caveman Communication'. These instructions are designed to be persistent and enforced across future workspace sessions.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting external workspace data to populate instruction files.
Ingestion points: Folder names from .code-workspace files and filesystem directory structures are read to build the Project Matrix.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are wrapped around project names when they are written to AGENTS.md.
Capability inventory: The agent identifies file structures and tech stacks and writes markdown files to the local filesystem.
Sanitization: Project identifiers ('codes') are normalized to uppercase alphanumeric characters with hyphens, which provides some sanitization, but full project names are inserted with only minor cleanup (trimming and icon removal).

agents-md