research-idea-explorer

Research Idea Explorer Skill

Use this skill when the task is to generate, refine, or compare research ideas across disciplines.

Default policy

If the machine has the packaged CLI installed, prefer calling:

• research-idea-explorer
• or rie

Use the repo-local fallback python -m research_idea_explorer.cli only when the packaged command is unavailable but the repository is present in the workspace.

Default to literature-first for every non-pure-functional invocation.

That means:

• If the user asks to generate, refine, compare, extend, red-team, or continue research ideas, search literature first.
• Treat literature retrieval as the default prelude, not an optional extra, unless the user explicitly says not to use it.
• Keep hybrid retrieval as the default strategy unless the user explicitly prefers lexical, embedding, or graph.
• Only skip retrieval for pure utility actions such as:
  • formatting an existing result
  • inspecting graph, summary, or neighbors
  • writing feedback
  • explaining the output schema or command usage

The workflow is:

1. Start with a literature-guided single-pass divergence across multiple research moves. The built-in persona set remains an internal way to keep those moves orthogonal, but it should not dominate the user-facing framing.
2. Convert only the strongest brainstorm seeds into the shared internal schema: Object, Puzzle, Claim, Contrast, Evidence, Scope, Stakes.
3. Reject ideas that are obvious near-duplicates or amount to a thin "method + topic" pairing.
4. If the user asks to deepen, branch, compare, or continue from an accepted direction, run a second literature-guided mutation round around the retained frontier families.
5. Push each retained idea one step further by adding: data source, minimal runnable design, likely failure mode, and why the question matters now.
6. When available, use the persistent memory graph so the session does not keep revisiting the same research neighborhoods.