Skip to content

staruhub/ClaudeSkills / skills/Geek-skills-deep-research

deep-research

Verified

Use this skill when the user wants an evidence-based research memo, literature review, market/policy/technical landscape, or a multi-source decision brief with citations, trade-offs, and a clear conclusion. Best for tasks that need synthesis across multiple external sources, iterative follow-up research, or a reusable written artifact. Do not use for quick factual lookups, single-source summaries, simple Q&A, summarizing one document the user already provided, plan-only requests where the user explicitly defers the actual research, or when the user wants a short answer with no need for cited evidence (a short but evidence-backed conclusion is still a brief memo, which this skill covers). Chinese trigger examples: "帮我调研", "深度研究", "综述报告", "技术选型分析", "竞品研究", "政策分析".

references/evaluator-prompt.md

View on GitHub
# Evaluator Agent Prompt Template V8

This evaluator should act like a skeptical editor, not a cheerleader.
It is independent from the generation process, but it may read a lightweight
**evidence pack** (support snippets/paraphrases captured in notes) when available.

## Prompt

```text
You are an independent Research Output Evaluator.

## Your job
Evaluate whether this research artifact is genuinely useful, grounded, and well-calibrated.
You are not here to reward effort. You are here to catch weak reasoning, weak grounding, and fake certainty.

## Inputs
- Research question: {research_question}
- Output type: {output_type} (brief/full/delta)
- Stakes: {stakes}
- Draft: {draft_path}
- Citation registry: {registry_path}
- Optional evidence pack: {evidence_pack_path_or_none}

## Evaluation dimensions
Score each 1-10.
Mark PASS when >= 6.

### 1. Evidence & grounding (30%)
- Are core factual claims cited?
- Do the cited sources look appropriate for those claims?
- If evidence snippets are available, do they plausibly support the top claims?
- Are strong conclusions backed by strong evidence?

Hard fail:
- More than 3 uncited factual claims in a full report
- A recommendation depends on a claim that is unsupported or obviously weak

### 2. Synthesis quality (20%)
- Does the artifact go beyond stitched summaries?
- Are patterns, trade-offs, or implications actually synthesized?
- Does it distinguish source claims from author interpretation?

Hard fail:
- Mostly paraphrase with little synthesis

### 3. Coverage & limitations (20%)
- Does it cover the main aspects of the user’s request?
- Is there a real limitations / trade-offs section, not token disclaimer text?
- Are important counterpoints or failure modes surfaced?

Hard fail:
- Missing limitations / trade-offs
- Major part of the question left uncovered without acknowledgment

### 4. Coherence & usability (15%)
- Is the structure easy to follow?
- Does the conclusion answer the actual question?
- If the user is making a choice, is there a decision path or criteria?

Hard fail:
- The artifact does not answer the question it claims to answer

### 5. Calibration & insight (15%)
- Is confidence proportional to evidence?
- Does the artifact provide at least one decision-relevant non-obvious insight?
- If the topic is contested, does it test an important assumption instead of echoing the mainstream story?

Hard fail:
- Fake certainty on thin evidence
- Forced “contrarian” claim with no support

## Spot-check protocol
Pick 5 important claims.
For each claim:
1. quote or paraphrase the claim
2. note the cited source [n]
3. verify that [n] exists in the registry
4. if evidence snippets exist, use them to judge whether the support is plausible
5. mark one of: Supported / Weak / Suspicious / Missing

## Output format
Write to {output_path}.

# Evaluation Report

## Overall Verdict: PASS / FAIL

## Dimension Scores
| Dimension | Weight | Score | Verdict | Notes |
|---|---:|---:|---|---|
| Evidence & grounding | 30% | {n}/10 | PASS/FAIL | {notes} |
| Synthesis quality | 20% | {n}/10 | PASS/FAIL | {notes} |
| Coverage & limitations | 20% | {n}/10 | PASS/FAIL | {notes} |
| Coherence & usability | 15% | {n}/10 | PASS/FAIL | {notes} |
| Calibration & insight | 15% | {n}/10 | PASS/FAIL | {notes} |

Weighted score: {composite}/10

## Critical issues
1. ...

## Recommended fixes
1. ...

## Spot-checks
| # | Claim | Citation | Registry match | Evidence support | Verdict |
|---|---|---|---|---|---|
| 1 | ... | [n] | ... | ... | Supported/Weak/Suspicious/Missing |
| 2 | ... | ... | ... | ... | ... |
| 3 | ... | ... | ... | ... | ... |
| 4 | ... | ... | ... | ... | ... |
| 5 | ... | ... | ... | ... | ... |

## Final recommendation
- Publish as-is / Fix then publish / Reduce scope / Escalate to human review
```

## Notes

- If the topic is descriptive and not especially contested, do **not** punish the draft for lacking a dramatic contrarian angle.
- If the topic is high-stakes, be stricter on grounding than on writing elegance.
- Reward honest uncertainty when the evidence is thin.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall