The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill directs users to clone an external, untrusted repository (github.com/alirezarezvani/claude-skills.git) and execute scripts from that repository, such as ./scripts/convert.sh. This pattern allows the execution of arbitrary code from a source outside of the reviewed files.
[EXTERNAL_DOWNLOADS]: The installation instructions rely on a personal GitHub repository that does not belong to the skill's stated author (Roger-235), which poses a supply chain risk. Additionally, it references an external installation via clawhub install.
[COMMAND_EXECUTION]: The skill utilizes shell commands to invoke Python scripts (scripts/extract_citations.py, scripts/format_summary.py) to process user-provided documents. Invoking local scripts on untrusted data creates a potential surface for command injection if inputs are not properly sanitized.
[PROMPT_INJECTION]: The skill is designed to summarize untrusted external documents (e.g., research papers, web articles) but lacks boundary markers or defensive instructions to protect against indirect prompt injection.
Ingestion points: Untrusted user-provided document files, academic papers, and web articles (e.g., document.txt, paper.txt).
Boundary markers: Absent; the workflow does not include delimiters or instructions to disregard embedded commands in the source material.
Capability inventory: Shell execution of conversion scripts (./scripts/convert.sh) and Python-based data processing utilities (scripts/extract_citations.py, scripts/format_summary.py).
Sanitization: Absent; no evidence of input validation, escaping, or content filtering for the data processed by the scripts.

research-summarizer