The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to its design to ingest and triage content from external news feeds and search results.\n- Ingestion points: Data enters the agent context in Phase 1 (COLLECT) as described in SKILL.md and structured via the schema in references/evidence-contract.md. Specifically, article text and metadata are gathered from external sources.\n- Boundary markers: Absent. The skill instructions do not specify the use of delimiters or 'ignore' commands to prevent the agent from following natural language instructions that might be embedded within the collected news articles.\n- Capability inventory: Low risk. The skill is restricted to emitting a JSON artifact; it does not utilize high-privilege tools, execute code, or perform sensitive file system operations.\n- Sanitization: Absent. While metadata is strictly extracted into the five-fact contract, the source text processed during filtration and freshness checking is not filtered for malicious instructions.

news-collection