You are a senior research lab director reviewing a researcher's entire portfolio of AI-driven research quests. Your job is to write a markdown report that surfaces themes, near-duplicates, gaps, and c
${generated_at}
Explore
130,009 skills indexed with the new KISS metadata standard.
${generated_at}
Your job: produce a structural plan for the experiment script that the next stage will fill in. Crucially, you do NOT write the function bodies yet — you map out the skeleton, name the constants with their sources, and lock the data flow. A clean outline lets the next stage spend its thinking budget
A prior stage produced a structural outline for the experiment. Your job: fill in every function body in the scaffold. You do NOT change function signatures, you do NOT rewrite the RESULT_JSON contract, you do NOT alter the imports or the figures-dir setup. The outline locked those decisions; your o
$design_block
brainstormed for the same topic. Pick which is more promising as the
$topic
$topic
$previous_code
This digest covers: **${window_from}** → **${window_to}**.
$topic
$topic
You are NOT running a simulation. You are NOT inventing data. You are summarizing what the user actually dropped into the data dir.
This is a Chain-of-Verification pass: the first pass is fast but tends to over-claim (a topically-related paper gets called *supporting* when its abstract is too thin to actually weigh in). The verification pass catches that by forcing the model to articulate WHY a claim of support or conflict holds
$topic
- Quest ID: `${quest_id}`
$topic
Given only the topic below, your job is to produce a short structured
$clarify_block
---
Scans newly written notes for wikilink opportunities and updates existing project notes with links to new notes.
Parses natural language research prompts into structured execution plans with topic detection, mode routing, and usage estimation.
Scans completed batch results for research leads, scores them by novelty and relevance, and proposes threads for follow-up research.
Searches the web for relevant sources on a research topic, prioritizing primary sources. Selects sources by tier and depth — 5-25+ URLs depending on the topic's depth profile.
Between-hop reasoning for the research pipeline. Computes confidence, picks the next hop pattern, scores candidate hops, and decides continue/stop/replan.