The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied paper drafts to perform its core review function, which creates a standard surface for indirect prompt injection if the content contains hidden instructions.
Ingestion points: The fiveAspectReview function in scripts/five_aspect_review.js accepts the user's paper draft as draftText for analysis.
Boundary markers: The skill uses a --- DRAFT --- header in dispatchAspect to separate task instructions from the ingested user content.
Capability inventory: The skill is limited to read_file, edit_file, write_file, and think_tool capabilities.
Sanitization: Ingested content is interpolated into analysis prompts without explicit sanitization or escaping of embedded instructions.

paper-review