Idea default plugin task - robbiemu/aclarai GitHub Wiki

🟢 Implement Fallback Plugin (LLM Agent)

Goal: Ensure that aclarai can handle any unrecognized or irregular input file by using a last-resort plugin that invokes an LLM agent to convert the content into Tier 1 Markdown format.

Task: Implement Fallback Plugin as a Conversion Agent

Description: Create a default plugin that always accepts the input, but delegates interpretation and formatting to an LLM-powered agent. This agent analyzes unstructured input and attempts to extract one or more conversations, converting each to standard aclarai Tier 1 Markdown format.

Agent Responsibilities:

  • Determine if the input contains a conversation.

    • If none, return nothing (plugin skips file).
    • If multiple conversations are found, split and return each separately.

  • For each conversation:

    • Format it as Markdown with speaker: text utterances.
    • Provide metadata: title, participants, message_count, etc.
    • Include a plugin_metadata field noting LLM inference was used.

Plugin Behavior:

  • Always returns True from can_accept(...)
  • Calls the agent in convert(...)
  • Wraps each result as a MarkdownOutput, and passes through ensure_defaults(...)
  • Skips file if agent returns None

Why this matters: This ensures aclarai can gracefully handle unstructured input, pre-formatted Markdown, or obscure chat logs without needing a format-specific plugin. It also provides robust onboarding for messy data during early experimentation.