| name | ship |
| description | Generates research or project ideas from papers reviewed so far. Triggers on "what could I do with this", "next research direction", "ideas based on this paper". |
ship — ideation
Process
- Ground every idea in a specific gap. Derive ideas from a concrete limitation, open problem, or unexplored combination found in the papers already reviewed — not from generic brainstorming disconnected from the literature.
- Rate feasibility realistically. The user does online-search-only research with no assumed compute cluster — weight ideas toward what's actually executable (small-scale experiments, analysis, surveys, applying an existing method to a new domain) over ideas requiring large-scale pretraining.
- Rate novelty honestly. Check whether the "gap" has already been addressed by other work found during literature_review — an idea that ignores existing work in the same session is not novel, it's an oversight.
- Score each idea on feasibility and novelty (e.g. high/medium/low each) so the user can prioritize.
- Write to
_ideas/ as a dated idea note. Each idea must cite, via Markdown link citation, which paper(s) it builds on or reacts against.
Anti-Rationalization
- "This would be a cool project" — coolness isn't grounding; tie it back to a specific gap in a specific paper.
- "I don't need to check if this was already done" — skipping this check produces ideas the literature review already falsified; always cross-check against papers already found in the session.
- "Feasibility doesn't matter, it's just brainstorming" — an infeasible idea list wastes the user's time; rate feasibility honestly even in early ideation.
Evidence
An idea list is complete when: every idea cites the paper(s) it's grounded in, feasibility and novelty are rated per idea, and no idea duplicates something already covered by a reviewed paper without acknowledging it.
Red Flags
- An idea with no citation back to reviewed papers
- Ideas that ignore an existing solution already surfaced in the same session
- No feasibility/novelty rating, just a flat list
