The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a mechanism to scan and index metadata from other skills in the environment, creating a potential vector for indirect prompt injection.
Ingestion points: The script scripts/_auto_sync.py iterates through sibling directories, reading SKILL.md files and parsing their YAML frontmatter (specifically name, description, and version) into a Python dictionary.
Boundary markers: There are no explicit boundary markers or sanitization routines for the metadata extracted from external SKILL.md files before it is processed.
Capability inventory: The skill includes file-writing capabilities through scripts/write_setup.py (which writes project preferences to .heyeddi/stack.json) and scripts/_auto_sync.py (which generates skills-index.json and skills-index.md).
Sanitization: While descriptions are truncated to 240 characters in _auto_sync.py, the content is not sanitized for markdown or control sequences before being written into the project's index files, which the agent is subsequently instructed to read for context.

heyeddi-setup