The Agent Skills Directory

[SAFE]: The skill is entirely instructional and composed of markdown text. It contains no executable scripts (Python, JavaScript, shell) or binary files, eliminating the possibility of direct local code execution or privilege escalation.
[EXTERNAL_DOWNLOADS]: The documentation references well-known and trusted AI services, including Anthropic's Claude, OpenAI's ChatGPT, and Google Gemini, as suggested tools for the research process. These references are for legitimate user-driven workflows and do not involve unauthorized remote fetching or hidden script downloads.
[DATA_EXFILTRATION]: There is no evidence of credential harvesting, hardcoded secrets, or unauthorized network operations. The skill focuses on processing public information and explicitly warns against using private or paywalled sources.
[PROMPT_INJECTION]: The skill outlines a methodology for processing external data into prompts. While the ingestion of untrusted internet content technically creates a surface for indirect prompt injection, the skill specifically mandates a human verification step (Step 4) to check source validity and provides specific boundary markers in its prompt template to separate user instructions from research context.
[SAFE]: Metadata and instructional content are consistent with the stated purpose of professional research. No obfuscation, deceptive language, or persistence mechanisms were detected.

deep-research