The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts to perform novelty assessments and literature searches.
Evidence: Running novelty_check.py and search_semantic_scholar.py located within local environment directories (~/.claude/skills/).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text descriptions of research ideas, which creates an indirect prompt injection surface if user-supplied descriptions contain hidden instructions.
Ingestion points: External research idea inputs enter via $0 or the --idea argument in SKILL.md and are interpolated into {idea_description} within references/assessment-prompts.md.
Boundary markers: Absent; there are no specialized delimiters or structural instructions to prevent the model from obeying instructions wrapped inside a malicious research proposal input.
Capability inventory: The skill interacts with local command lines by executing automated check scripts and query generation scripts.
Sanitization: Absent; input text is passed into the prompt templates and command line arguments without escaping or verification mechanisms.

novelty-assessment