The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from untrusted external sources, including scholarly papers, web pages, and Git repository READMEs.\n
Ingestion points: Artifacts are discovered via scholarly APIs in scripts/search_sources.py and downloaded in scripts/fetch_artifacts.py.\n
Boundary markers: The skill contains explicit defensive instructions in SKILL.md and references/reading-a-paper.md advising the agent to treat fetched content as data rather than instructions and to ignore any 'ignore previous instructions' patterns found in artifacts.\n
Capability inventory: The skill can perform network operations (scripts/common.py), write files to the local filesystem (various scripts), and execute shell commands via git (scripts/fetch_artifacts.py).\n
Sanitization: The skill relies on structural parsing (JSON/XML) and manual validation steps (verify_locators.py) but lacks automated sanitization of the natural language content before processing.\n- [EXTERNAL_DOWNLOADS]: The skill fetches data from several external scholarly APIs and clones remote Git repositories.\n
Scholarly APIs: It communicates with export.arxiv.org, api.semanticscholar.org, api.openalex.org, and api.crossref.org. These are well-known scholarly services.\n
Repository Cloning: The scripts/fetch_artifacts.py script performs shallow clones of Git repositories provided via the --repo argument.\n- [COMMAND_EXECUTION]: The skill uses the subprocess module to execute git clone and git rev-parse commands in scripts/fetch_artifacts.py for the purpose of acquiring research artifacts.

literature-survey