The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of instructions attempting to bypass safety filters, override system guidelines, or reveal internal prompts was found. The skill uses professional instructional language and structural checkpoints to guide the agent's behavior.
[DATA_EXFILTRATION]: The skill does not contain any commands for network exfiltration or access to sensitive local file paths (e.g., credentials, SSH keys). It mentions a project-specific log file (.research/decision-log.yaml) for prerequisite verification, which is within the expected scope of the research activity.
[REMOTE_CODE_EXECUTION]: There are no patterns involving the download or execution of remote scripts. While the skill recommends external transcription software like Otter.ai or Descript, these are provided as informational references for the user and are not executed by the agent.
[OBFUSCATION]: No hidden content, encoded strings (such as Base64), zero-width characters, or homoglyph substitutions were detected. The markdown content is clear and legible.
[COMMAND_EXECUTION]: The skill does not use shell commands, subprocesses, or privilege escalation techniques (e.g., sudo). It uses a custom tool-like syntax for prerequisites (diverga_check_prerequisites), which appears to be a standard Model Context Protocol (MCP) call for the intended platform environment.
[INDIRECT_PROMPT_INJECTION]: While the skill is designed to process user-provided research data and questions, it lacks dangerous capabilities (such as file writing or network access) that would allow for exploitation via poisoned input.

d2