The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest external data from web searches regarding market sentiment, search trends, and media narratives to calculate bubble scores. This creates a surface for potential indirect prompt injection from malicious web content. 1. Ingestion points: Web search results for metrics like Google Trends, social penetration reports, and media narratives (SKILL.md, references/implementation_guide.md). 2. Boundary markers: The skill mandates a 'Confirmation Bias Prevention Checklist' and strict mechanical scoring to prevent the agent from over-weighting external subjective content. 3. Capability inventory: The agent uses web_search to collect data and executes the bubble_scorer.py script to finalize assessments. 4. Sanitization: The instructions include strict validation criteria, such as requiring specific dates, names, and a minimum of three independent sources for qualitative adjustments.
[DYNAMIC_EXECUTION]: The skill's test infrastructure (scripts/tests/test_skill_contract.py) uses Python's importlib library to dynamically load and execute the bubble_scorer.py module. While this is a standard practice for local script validation, it represents a dynamic code loading pattern.

us-market-bubble-detector