The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it is designed to ingest and process various external project artifacts.
Ingestion points: According to SKILL.md, the agent reads multiple files from the .byte-os/ directory, including PRODUCT_SPEC.md, TECH_SPEC.md, and roadmap files.
Boundary markers: Absent. The instructions do not define delimiters or specific warnings to prevent the agent from following instructions potentially embedded within the analyzed documents.
Capability inventory: The skill involves writing to the local file system, specifically creating new review documents in .byte-os/reviews/ and modifying STATUS.md.
Sanitization: Absent. No mention of content validation, escaping, or filtering of the ingested file data is present in the skill instructions.

byte-review