The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The Python script scripts/run.py is designed to download files and scrape HTML tables from external URLs provided via a configuration file (data_requirements.json). By default, it includes references to example.com and wikipedia.org.
[COMMAND_EXECUTION]: The SKILL.md file contains instructions requiring the agent to execute specific local Python scripts (e.g., workflow_guard.py, update_workflow_memory.py) located in sister directories (.claude/skills/paper-workflow-orchestrator/ and .claude/skills/context-memory-keeper/). These commands are used to maintain state and enforce workflow constraints across a multi-skill environment.
[DATA_EXFILTRATION]: The skill uses the requests library to fetch data. While this involves network operations, the primary intent is data ingestion (downloading to crawled_data/) rather than exfiltration of sensitive local information.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from external URLs (HTML tables and CSVs) and processes them into the agent's context.
Ingestion points: Data is fetched from remote URLs in scripts/run.py and saved to crawled_data/.
Boundary markers: None identified in the prompt templates or data processing logic.
Capability inventory: The skill has the capability to write files, perform network requests, and execute local scripts as part of its workflow.
Sanitization: There is no explicit sanitization or filtering of the content retrieved from external sources before it is processed or presented to the user/agent.

authoritative-data-harvester