The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs and uses standard Python libraries (openai, anthropic, google-genai) and communicates with official API endpoints for OpenAI, Anthropic, Google, and Perplexity AI. These are well-known technology services and are considered safe sources for their respective functions.
[COMMAND_EXECUTION]: The skill instructions in SKILL.md guide the agent to execute a sequence of local Python scripts (probe_llms.py, analyze_results.py, build_report.py) to perform the audit pipeline. These scripts manage data flow from API ingestion to final report generation.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it analyzes data retrieved from external LLMs which may include information from untrusted sources.
Ingestion points: scripts/probe_llms.py (ingests responses from OpenAI, Anthropic, Google, and Perplexity APIs).
Boundary markers: Absent; the system prompt and templates do not use specific delimiters to separate untrusted content from the analysis instructions.
Capability inventory: The skill has capabilities for network operations (API calls in probe_llms.py) and file system writes (writing analysis and reports in data/ and report/).
Sanitization: The skill implements basic length capping and regex-based extraction but does not sanitize the incoming text for potentially malicious instructions that could influence the agent's report generation logic.

geo-gap-fixer