菜单
Use when a researcher is stuck, unsure of the next step, feels lost in the middle of a project, or asks any variant of "what should I do next?", "where am I?", "what now?", "I'm stuck" — diagnoses research state and recommends which Eureka skill to invoke next
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use during or after manuscript writing to verify every quantitative claim traces to actual data, every figure is script-generated, and all results including negative ones are reported
Use when starting any conversation involving research, hypotheses, experiments, data analysis, scientific claims, or manuscript writing — establishes how to find and use Eureka research skills
Use when a research design has been approved and needs to be broken into executable experiment steps — the research equivalent of an implementation plan
Use when creating, updating, or reviewing a research figure (plot, chart, brain map, heatmap, schematic) — guides chart-type selection, typography, colorblind-safe palette, layout, and journal-specific export specs. Dispatches a figure-reviewer subagent after rendering. Complements claims-audit (which checks figure integrity) by checking figure design.
Use when running any experiment or analysis — hypothesis and analysis plan must be registered before seeing results, the scientific equivalent of test-driven development
Use when writing a research manuscript — guides section-by-section writing with prerequisite gates, citation discipline, number traceability, and per-section subagent review. Activated after experiment results are reviewed and before claims-audit.
| name | whats-next |
| description | Use when a researcher is stuck, unsure of the next step, feels lost in the middle of a project, or asks any variant of "what should I do next?", "where am I?", "what now?", "I'm stuck" — diagnoses research state and recommends which Eureka skill to invoke next |
Researchers get stuck. Not because they're lazy or incompetent — because the research lifecycle has many branches, every project has different state, and "what should I do now?" is a hard question even for experienced PIs.
This skill diagnoses where you are in the research lifecycle, then recommends the next Eureka skill to invoke. It does NOT do the work itself. It is a router — it figures out which specialist to call.
Core principle: When stuck, the answer is almost never "try harder at what you're doing." The answer is usually "you're in a different phase than you think — switch tools."
This skill has two modes:
User doesn't know where they are in the research lifecycle, or needs help picking the next skill. Scan state → diagnose phase → recommend specialist skill. This is the existing workflow (Steps 1-6 below).
User returns after a break and wants to continue where they left off. Trigger phrases: "resume", "continue", "last session", "where was I", "어디서부터", "이어서".
Resume workflow:
docs/eureka/journal/ (sort by filename YYYY-MM-DD.md). If no journal entry exists, fall back to Diagnostic mode.git log --oneline -20 for context on what was worked on.docs/eureka/ subdirectories for most-recently-modified artifacts:
designs/ — any new design docs?registrations/ — any new or amended registrations? (check INDEX.md)plans/ — any experiment plans in progress?audits/ — recent claims-audit runs?reviews/ — recent research-reviewer runs?novelty-audits/ — recent novelty audits?verifications/ — recent verification runs?Decision rule: if the journal entry clearly identifies the current phase AND there are no stale artifacts suggesting the state has drifted, use resume mode. If the journal entry is ambiguous or more than 2 weeks old, fall through to diagnostic mode.
Use this skill when:
Do NOT use this skill when:
This skill never executes the actual research work. It always ends with a recommendation to invoke another Eureka skill. If you find yourself starting to brainstorm a hypothesis, troubleshoot a result, or write a manuscript while in this skill — STOP. You skipped the handoff.
The job of whats-next is to answer one question: "Which Eureka skill should I invoke right now?"
You MUST create a task for each of these and complete them in order:
Before asking anything, scan the project files to gather objective evidence:
| Look for | What it tells you |
|---|---|
docs/eureka/designs/*.md | A research design exists → past brainstorming phase |
docs/eureka/registrations/*.md | Hypothesis is registered → past hypothesis-first phase |
docs/eureka/plans/*.md | Experiment plan exists → past experiment-design phase |
results/ directory with files | Experiments have been run |
docs/eureka/records/experiment-log.md | Structured experiment tracking exists |
docs/eureka/audits/*.md | Claims have been audited |
docs/eureka/journal/*.md (most recent) | Narrative context from last session — decisions, failures, blockers, and the "Next session starts with" hint. This is your strongest continuity signal. |
Recent git log (last 5–10 commits) | What the user actually did most recently |
*.tex, manuscript/, paper/ | Manuscript writing is in progress |
notes/ or other free-form files | Additional free-form thinking that may reveal context |
If the project does not use Eureka conventions yet (no docs/eureka/ structure), the user is likely at the very beginning. That is its own diagnosis — the answer will probably be eureka:research-brainstorming.
Look at the latest commits, the newest files in results/, the most recently modified document. State what you found:
"I see your most recent activity: 3 baseline experiment results in
results/run_20260411_*/, last commit wasRun baseline modelfrom 2 days ago. The design doc exists but no review report yet."
This is not an interpretation — it is a factual report. The user can correct it if you read the state wrong.
If docs/eureka/journal/ exists, read the most recent entry (latest date file, last ## session block within it). The Next session starts with field is your strongest single signal for what to recommend — it is a message from past-user to future-user specifically about the next action. Quote it back to the user:
"Your last journal entry (2026-03-22, 18:00 session) says next session starts with: 'Re-run Model A with per-subject normalization'. Is that still the right starting point, or has anything changed since?"
The journal also surfaces Blockers and Failed attempts from prior sessions, which tell you what to avoid recommending.
Ask one at a time. Stop as soon as you have enough information to recommend.
The four core questions (skip any that are already answered by Step 1 or by the conversation):
Optional follow-ups when needed:
Based on Steps 1–3, identify which phase the user is in. Use this map:
| State | Phase | Recommend |
|---|---|---|
| No design doc, no research question at all — only keywords, a dataset, or vague interest (e.g., "something with EEG") | Pre-ideation | eureka:research-ideation |
| No design doc, has a research question but it needs refinement (e.g., "Does X cause Y?") | Pre-design | eureka:research-brainstorming |
| Design exists, no registration, no analysis | Pre-registration | eureka:hypothesis-first |
| Hypothesis registered, no experiment plan | Pre-execution planning | eureka:experiment-design |
| Experiment plan exists, no results | Execution | Just run the experiments — no skill needed yet |
| Some results, but mixed/unclear/unexpected | Investigation | eureka:systematic-troubleshooting (if technical) OR eureka:requesting-research-review (if interpretive) |
| Results complete, not yet evaluated | Phase review | eureka:requesting-research-review |
| Review passed, starting to write | Manuscript drafting | eureka:claims-audit (use during writing, not just at the end) |
| Manuscript drafted | Pre-publication audit | eureka:claims-audit then eureka:verification-before-publication |
| Verification passed | Submission decision | eureka:submission-readiness |
| Pivoting / lost direction | Re-design | eureka:research-brainstorming (start over with new question) |
If the user's state spans multiple phases, pick the earliest unfinished one. You cannot skip Phase N to get to Phase N+1.
Discriminating rule for Pre-ideation vs Pre-design: If the user can state the question as "Does/Is/Can [X] [verb] [Y]?", they have a formed question → Pre-design. If they cannot, they are in Pre-ideation.
Present the recommendations in this format:
You are currently in: [phase name]
Most likely next actions:
1. [Recommended] Invoke `eureka:[skill-name]` — [one-sentence rationale]
2. Alternative: Invoke `eureka:[other-skill]` — [when this is the better choice]
3. Optional: [non-skill action, e.g., "wait for collaborator response"]
Which one?
Lead with one clear recommendation. The alternatives exist for cases where the user has context you don't.
After the user picks, invoke the chosen skill via the Skill tool. Do NOT try to do that skill's job inside whats-next.
If the user picks the recommended skill, just invoke it. If they pick an alternative, invoke that one — they have context you don't.
After the hand-off, gently remind the user:
"At the end of this session,
eureka:research-journalcan capture today's decisions and what to start with next time."
This is a soft reminder, not a requirement. The user decides whether to journal at session end.
Total response length: short. This is a triage skill, not a coaching session. Aim for:
Long outputs defeat the purpose. The user is stuck. They need a direction, not a lecture.
| What you might do wrong | What to do instead |
|---|---|
| Start brainstorming the next experiment | Stop. Recommend eureka:research-brainstorming and hand off. |
| Diagnose technical bugs in the data | Stop. Recommend eureka:systematic-troubleshooting and hand off. |
| Critique the user's research direction | Stop. The right skill for evaluation is eureka:requesting-research-review. |
| Write a long motivational/coaching response | Triage is fast and factual. Save the empathy for a different conversation. |
| Recommend more than 3 actions | Decision paralysis is part of the problem. Narrow it down. |
| Skip the project state scan | Without scanning, your recommendation is a guess. Always scan first. |
| Recommend a skill the user has clearly already done | Read the project state. If they registered a hypothesis already, do not tell them to register one. |
| Treat this as a brainstorming session | This skill is a router. The brainstorming happens in eureka:research-brainstorming. |
These thoughts mean STOP — you're slipping out of triage mode:
systematic-troubleshooting.claims-audit.These are the most frequent stuck states researchers experience and the typical correct routing:
| Stuck state | Likely phase | Skill to recommend |
|---|---|---|
| "Baseline done, results are mixed" | Phase review | requesting-research-review to evaluate rigor before drawing conclusions |
| "I have an idea but don't know how to test it" | Pre-design | research-brainstorming |
| "Experiment crashed and I don't know why" | Execution → Investigation | systematic-troubleshooting |
| "The numbers look right but feel wrong" | Investigation | systematic-troubleshooting (Phase 1: investigate) |
| "Reviewer said X, not sure how to respond" | Post-review | receiving-research-review |
| "I think I'm done but not sure" | Pre-publication | verification-before-publication |
| "Should I keep going or pivot?" | Decision | submission-readiness (the four-option gate) |
| "I have results but no plan to write them up" | Drafting | Start writing, then claims-audit |
| "I haven't touched this in weeks, where was I?" | Any | Scan state, recommend the next gate |
| "I keep going in circles on the same analysis" | Investigation OR re-design | systematic-troubleshooting first; if 3+ attempts failed, escalate to research-brainstorming |
| "I have data but don't know what to study" | Pre-ideation | research-ideation |
| "What interesting questions are there in this field?" | Pre-ideation | research-ideation |
| "The figure looks wrong / won't pass journal review" | Writing / submission prep | figure-design |
| "I need to make a figure for Results" | Writing | figure-design (after results are finalized) |
| "Results are technically fine but the paper feels flat" | Writing / pre-submission | manuscript-writing narrative-arc lock step (Discovery-Adjusted Framing) with docs/references/narrative-guide.md section "Discovery-Adjusted Framing" |
| "Not sure if this is Nature-level or specialty-journal-level" | Submission prep | submission-readiness venue-framing check with narrative-guide.md section "Venue-specific altitude tuning" |
| "The paper is rigorous but I'm not sure it's still novel" / "worried about preemption" | Pre-submission | novelty-competitive-audit (blocks submission-readiness until PASS) |
| "Reviewer/editor surfaced a preempt paper; what do I do?" | Pre-revision / pre-resubmission | novelty-competitive-audit first, then manuscript-writing narrative-arc lock re-fire if altitude needs adjustment |
eureka:using-eureka when the user expresses being stuck or asks "what next?"FLEXIBLE — The diagnostic conversation adapts to context. The order and depth of questions can change based on what the project state already reveals. The constant is the structure: scan → identify → diagnose → map → recommend → hand off.
This is a router skill. Routers do not solve problems — they make sure the right specialist gets called. Discipline here means resisting the urge to "just help" inside the router itself.