The Agent Skills Directory

[SAFE]: The skill's primary function is to transform research paper data (CSV, abstracts, full text) into structured notes (JSONL). The workflow involves reading local workspace files and performing deterministic text extraction.
[COMMAND_EXECUTION]: The skill uses a scripts/run.py file to automate the scaffolding of notes. Technical analysis of this script confirms it only performs local file I/O (reading core_set.csv, writing paper_notes.jsonl) and string manipulation using standard Python libraries. There are no calls to subprocess, os.system, or eval that would allow arbitrary command execution.
[DATA_EXFILTRATION]: The skill documentation explicitly states 'Network: none'. Review of scripts/run.py confirms there are no network-related imports (e.g., requests, urllib, socket) or operations that could transmit data externally.
[CREDENTIALS_UNSAFE]: No hardcoded API keys, tokens, or secrets were found in the codebase. The skill operates on academic paper metadata which is generally public data.
[PROMPT_INJECTION]: The skill includes 'Role cards' and 'Guardrails' in its instructions. These are used to guide the LLM in extracting specific, high-quality information (e.g., 'Avoid generic summary boilerplate'). These are benign instructional constraints and do not attempt to bypass system safety filters.
[EXTERNAL_DOWNLOADS]: The skill does not perform any external package installations or remote script downloads. It relies on a local tooling.common module which is expected in this agent environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (paper abstracts and full texts). While this is an attack surface, the skill implements 'Source Text Hygiene' through regex patterns in assets/source_text_hygiene.json to sanitize and filter author-centric or promotional language, which serves as a basic form of data validation and sanitization.

paper-notes