メニュー
Use when user wants to find gaps, evaluate ideas, or decide what's worth pursuing. Triggered by "what hasn't been tried", "is this a good idea", "find the gap".
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | idea-selection |
| description | Use when user wants to find gaps, evaluate ideas, or decide what's worth pursuing. Triggered by "what hasn't been tried", "is this a good idea", "find the gap". |
"The single most important skill is good taste in what problems are worth solving." — Nicholas Carlini
Check if a lab notebook exists:
ls .research-loop/sessions/
If yes, read the most recent one to understand context. If no, ask:
"What topic are we evaluating? (Or run
/explore <topic>first to build the foundation.)"
Spawn these 3 simultaneously:
Agent 1: What has NOT been tried in [topic]? Find explicit "open problem" or "future work" statements.
Agent 2: What methods from adjacent fields haven't been applied to [topic]?
Agent 3: What assumptions does the [topic] literature make that might be wrong?
Save raw results to lab_notebook.md:
## Gap Analysis (raw)
### Untried approaches
<agent 1 results>
### Adjacent field transfers
<agent 2 results>
### Questionable assumptions
<agent 3 results>
Then present top 3 gaps as options:
"I found 3 gaps worth considering:
1. [Gap name] — [one sentence]. Feasibility risk: [one sentence]. 2. [Gap name] — [one sentence]. Feasibility risk: [one sentence]. 3. [Gap name] — [one sentence]. Feasibility risk: [one sentence].
Which one resonates? Or a different angle?"
Append their choice:
## Selected Gap
Gap: <their choice>
Why they chose it: <their reasoning>
Ask one question at a time. Wait for each answer. Score internally (don't reveal yet).
Q1 — Taste:
"First — be honest: if someone solved this completely, would the field look meaningfully different? Or is this more of a nice-to-have?"
Append:
## Carlini Gate
### Taste (weight 0.30)
Question: Would solving this meaningfully change the field?
Researcher answer: <their answer>
Score: <0.0-1.0>
Reasoning: <why that score>
Q2 — Uniqueness:
"What can YOU specifically bring to this that others can't? Your background, timing, how you're framing it — what's your edge?"
Append:
### Uniqueness (weight 0.25)
Question: What is your comparative advantage?
Researcher answer: <their answer>
Score: <0.0-1.0>
Reasoning: <why that score>
Q3 — Impact:
"Write the best-case conclusion right now. If every experiment works perfectly — what does the paper say? Not 'X% improvement' — what changes about how people think?"
Append:
### Impact (weight 0.30)
Question: What is the best-case conclusion?
Researcher answer: <their answer>
Score: <0.0-1.0>
Reasoning: <why that score>
Q4 — Feasibility:
"Can you test the core claim with a single GPU in under a week? Describe the exact experiment."
Append:
### Feasibility (weight 0.15)
Question: What is the exact experiment?
Researcher answer: <their answer>
Score: <0.0-1.0>
Reasoning: <why that score>
Calculate and show:
Taste: X.XX × 0.30 = X.XX
Uniqueness: X.XX × 0.25 = X.XX
Impact: X.XX × 0.30 = X.XX
Feasibility: X.XX × 0.15 = X.XX
─────────────────────────────────
Overall: X.XX
Append to lab_notebook.md:
### Final Score
Taste: X.XX | Uniqueness: X.XX | Impact: X.XX | Feasibility: X.XX
Overall: X.XX
Verdict: <promising / conditional / skip>
Weakest axis: <which one and why>
If ≥ 0.70:
"Strong signal. Ready to run parallel discovery lanes? →
/discover"
If 0.50–0.69:
"Promising but [weakest axis] is the weak point. [One sentence on what would strengthen it.] Proceed anyway, or work on that first?"
If < 0.50:
"Honest take: not strong enough yet. Main problem is [weakest axis]. Want to try a different gap, or reframe the approach?"
Update lab_notebook.md status:
## Status
<date>: Carlini gate result: <verdict>. Next: <what to do>
Use when user mentions autonomous iteration, metric-driven optimization, $research-loop plan, $research-loop debug, $research-loop fix, $research-loop security, $research-loop ship, $research-loop scenario, $research-loop predict, $research-loop learn, $research-loop reason, $research-loop probe, or mentions "research-loop" with a goal/metric. Autonomous Goal-directed Iteration — apply Karpathy's autoresearch principles: modify, verify, keep/discard, repeat. Supports bounded mode via Iterations: N inline config.
Mandatory activation layer — loads on any conversation start. Establishes skill-loading protocol, Red Flags, priority rules, and HARD-GATE enforcement for all research-loop skills.
Run a thorough, multi-phase deep research investigation on a topic with subagent dispatch, provenance tracking, and integrity verification.
Experiment sandbox execution for Research Loop. Supports four modes: local (venv), Docker (isolated containers), SSH remote (GPU compute on servers), and Colab (Google Drive bridge). Provides experiment harness templates, code validation, metric collection, deterministic seeding, and compute budget enforcement. Use before running experiments generated by the paper-pipeline.
Use when user wants to explore a topic, find papers, map a field, or understand the research landscape. Not for explaining concepts — use learn for that.
Publication-quality figure generation for research papers. Decision agent selects figure type (code plot vs architecture diagram). Generates Matplotlib/Seaborn code for quantitative figures with iterative improvement loop. Style-matches conference templates (NeurIPS, ICML, ICLR). Use when the paper-pipeline reaches the figure generation phase, or when a user requests figures for an existing draft.