The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses persona-based instructions to guide the agent's reasoning. It does not contain commands to override safety filters, disregard previous instructions, or disclose system prompts. The persona is strictly bounded to economic analysis.
[DATA_EXFILTRATION]: There are no patterns indicating access to sensitive system files (e.g., .ssh, .aws) or hardcoded credentials. No unauthorized network operations are defined.
[OBFUSCATION]: No base64-encoded strings, zero-width characters, homoglyphs, or other techniques to hide malicious intent were found in any of the skill files.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute scripts from external sources. It relies entirely on static markdown instructions and internal reference files.
[COMMAND_EXECUTION]: No shell commands, privileged operations (sudo), or persistence mechanisms (cron, shell profiles) are present.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided economic scenarios. However, because it lacks capabilities for network access, file system modification, or process execution, the attack surface for indirect injection is not exploitable. It acts as a purely consultative analysis tool.
[DYNAMIC_CONTEXT_INJECTION]: The skill does not use the dynamic context injection syntax (!command) to execute shell commands at load time.

keynes-perspective