The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted source material (academic papers, web articles, PDFs) through its summarization and citation workflows. This creates a surface where an attacker-controlled document could contain hidden instructions attempting to influence the agent's summary or behavior.
Ingestion points: Workflow steps in SKILL.md ingest external text provided by the user for summarization, comparison, and citation extraction.
Boundary markers: The prompt instructions lack explicit delimiters or "ignore embedded instructions" warnings for the agent when processing external content.
Capability inventory: The skill uses Python scripts for formatting and extraction but does not have high-privilege capabilities such as network access or file-writing during runtime.
Sanitization: scripts/extract_citations.py uses regular expressions to extract specific structured patterns (DOIs, URLs), which provides a baseline level of data filtering.
[COMMAND_EXECUTION]: The skill utilizes local Python scripts to perform citation extraction and template generation.
Evidence: SKILL.md contains usage examples for python3 scripts/extract_citations.py and python3 scripts/format_summary.py. Analysis of the source code for these scripts shows they use standard libraries for text processing and do not engage in dangerous shell or system operations.
[EXTERNAL_DOWNLOADS]: The documentation references external installation sources and a third-party registry for non-standard deployment.
Evidence: Installation instructions in SKILL.md refer to github.com/alirezarezvani/claude-skills.git and the clawhub registry. While these are vendor-owned resources corresponding to the author, they constitute external dependencies from sources outside the pre-approved trusted vendor list.

research-summarizer