The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input (topic and domain) and interpolates it directly into a system-level prompt for the language model, creating a surface for prompt injection attacks.
Ingestion points: The --topic and --domain arguments in scripts/quick_research.py are the primary entry points for untrusted data.
Boundary markers: The prompt construction in create_research_prompt lacks delimiters or specific instructions to the model to ignore potential injection attempts within the topic string.
Capability inventory: The skill has the capability to write files to the user's home directory (~/research_briefs) and perform network operations via the Anthropic API and PubMed services.
Sanitization: No sanitization or validation is performed on the topic or domain strings before they are inserted into the f-string prompt.
[DYNAMIC_EXECUTION]: The script scripts/quick_research.py performs dynamic loading of a local module using a computed file path.
Evidence: The script uses sys.path.insert(0, str(Path(__file__).parent.parent.parent.parent.parent / "scripts")) to modify the Python search path at runtime. This allows it to import PubMedClient from a location relative to the script's directory, which is a form of dynamic loading from computed paths.

quick-topic-researcher