The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requests broad shell permissions via settings.local.json, including Bash(chmod +x:*), Bash(curl:*), and Bash(python:*). These permissions are utilized for environment initialization and for running the internal Python scripts provided with the skill.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external academic PDFs and bibliographic APIs to generate notes and synthesize research proposals.
Ingestion points: commands/read.md (reading PDF content) and commands/feed.md (fetching metadata from Semantic Scholar).
Boundary markers: The instructions for processing external content do not include explicit delimiters or warnings to the agent to ignore potentially malicious instructions embedded within the papers or metadata.
Capability inventory: The skill has the ability to write to the file system, perform network operations, and execute subprocesses through its utility scripts.
Sanitization: scripts/rename_pdfs.py applies regex filtering when generating filenames from metadata, but no explicit sanitization is used for text content interpolated into agent prompts during note generation.
[EXTERNAL_DOWNLOADS]: The skill fetches academic data from established and well-known services.
Evidence: Metadata queries are directed to api.semanticscholar.org (Semantic Scholar) and api.crossref.org (CrossRef).

litbase