The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions are purely task-oriented and do not contain any patterns intended to bypass safety filters, override system prompts, or induce unrestricted behavior. The instructional language used (e.g., "IMPORTANT", "ARCHITECTURE OVERVIEW") is benign and defines the operating logic for the agent.
[CREDENTIALS_UNSAFE]: No hardcoded API keys, tokens, or private secrets were found. The Python code snippets provided in the README and Usage Guide use standard library initializations with placeholders, which is a safe and recommended practice.
[DATA_EXFILTRATION]: There are no commands or network operations aimed at accessing sensitive local files (such as .ssh or .aws credentials) or sending them to external servers. The network operations described in the /research mode are clearly defined as legitimate web searches for competitive intelligence.
[OBFUSCATION]: The skill uses clear, human-readable markdown and YAML. No Base64, zero-width characters, homoglyphs, or steganographic patterns were detected in the text or URLs. All external links point to legitimate professional profiles or well-known service domains.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute remote scripts. It provides Python snippets as examples for users to integrate the skill into their own applications using the official Anthropic SDK, which is a trusted and standard resource.
[DYNAMIC_CONTEXT_INJECTION]: No use of the dynamic command execution syntax (!command) was found in the SKILL.md file or any referenced documents. The skill relies entirely on text-based prompt interpolation.

aeo-strategist