The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external implementation data (C# code, Unity assets, HTML sandboxes) and can modify the repository, creating a surface for indirect prompt injection.
Ingestion points: The skill reads arbitrary files from Assets/ and production/prototypes/ as specified in references/claude-skill.md (Phase 2).
Boundary markers: The instructions do not define explicit boundary markers or delimiters for the content being analyzed to prevent the agent from following instructions embedded in the code comments or prototype text.
Capability inventory: The skill has access to Read, Write, Edit, and Bash tools, allowing it to modify the codebase based on its analysis.
Sanitization: There is no automated sanitization of ingested content. While the skill encourages a collaborative protocol (Phases 3-6) that includes user review, the 'Codex Workflow' in SKILL.md suggests making changes directly when the request is clear, which increases the risk of acting on malicious instructions embedded in the source material without sufficient validation.

reverse-document