The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents an installation method that fetches a script from the vendor's official domain (https://paperclip.gxl.ai/install.sh) and pipes it directly into the Bash interpreter. It also supports downloading and installing Python wheels directly from the same domain.
[COMMAND_EXECUTION]: Instructions include sourcing local .env files using the shell . or source command to load environment variables. This executes the content of the file as shell code. The skill provides a warning to only source trusted, shell-compatible files.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted scientific literature and regulatory data from external repositories (PMC, arXiv, FDA, etc.) for analysis and extraction.
Ingestion points: External data enters the agent context through search, grep, lookup, and document reading commands documented in references/search-and-retrieval.md and references/cli-reference.md.
Boundary markers: SKILL.md explicitly instructs the agent to treat all service content and returned documentation as untrusted data that cannot override user instructions or authorize egress.
Capability inventory: The skill utilizes Bash, Read, and Write tools to interact with the Paperclip CLI, manage local files, and source environment configurations.
Sanitization: The instructions in references/map-reduce.md emphasize that extraction results are AI-generated and must be verified against primary evidence and specific line-pinned citations within the source documents.
[DYNAMIC_EXECUTION]: The skill uses map and reduce operations to perform LLM-based extraction, filtering, and synthesis of data from external documents. It also supports the execution of server-side routines and specialized workers for data processing.

paperclip