The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow for scanning and processing content from untrusted external sources, which creates a surface for indirect prompt injection attacks.
Ingestion points: The pipeline in SKILL.md includes scanning arXiv papers, technical blogs (e.g., from Anthropic, OpenAI, and DeepMind), and various industry reports.
Boundary markers: No explicit delimiters or boundary markers are defined in the documentation to separate external content from internal agent instructions.
Capability inventory: The agent is instructed to discover papers, extract findings, score applicability, and feed this data into a "research self-improvement loop."
Sanitization: There are no described steps for sanitizing, validating, or filtering the content ingested from the web before the agent processes it.
[NO_CODE]: The skill consists only of a SKILL.md file providing documentation and process descriptions, with no associated executable code or scripts.

paper-scanner