The Agent Skills Directory

[SAFE]: No security issues detected. The skill provides instructions for a business evaluation framework involving structured scoring of target markets.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages state by reading and writing to a local file named PROBLEM-SCORE.md. This is a documented behavior for persistence across user sessions and does not involve sensitive system paths or credential access.
[EXTERNAL_DOWNLOADS]: The skill instructions suggest performing lightweight research using web search tools to verify market data. This uses standard agent search capabilities for informational purposes and does not involve downloading executable code.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted user input regarding business ideas and processes it to generate scores. However, the skill implements a skeptical 'grilling' posture and structured output format which mitigates the risk of accidental instruction following from the input data.
Ingestion points: User-provided business idea descriptions and subsequent justifications during the scoring phase.
Boundary markers: The skill uses a structured Markdown table to delimit scores and justifications in PROBLEM-SCORE.md.
Capability inventory: File read/write access (limited to PROBLEM-SCORE.md) and web search capabilities.
Sanitization: No explicit sanitization of user input is described; the skill relies on the LLM's inherent reasoning and the framework's Skepticism rules.

asb-problem