notque/vexjoy-agent / skills/research/research
research
VerifiedResearch: structured investigation, fact-checking, explanation traces.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 8.0 KB |
| references/error-handling.md | markdown | 7.4 KB |
| references/preferred-patterns.md | markdown | 6.5 KB |
| references/research-coordinator/checklists/source-quality.txt | markdown | 14.8 KB |
| references/research-coordinator/delegation-patterns.md | markdown | 5.8 KB |
| references/research-coordinator/error-catalog.md | markdown | 4.4 KB |
| references/research-coordinator/examples/bad-research-plan.txt | markdown | 7.9 KB |
| references/research-coordinator/examples/good-research-plan.txt | markdown | 13.9 KB |
| references/research-coordinator/query-classification.md | markdown | 4.0 KB |
| references/research-coordinator/templates/subagent-instructions.txt | markdown | 17.3 KB |
| references/research-subagent-executor/source-quality-assessment.md | markdown | 3.5 KB |
| references/trace-schema.md | markdown | 7.4 KB |
references/research-coordinator/examples/bad-research-plan.txt
View on GitHub# Bad Research Plan Example
## Query
"Research the current state of Kubernetes cluster management in 2025: hosted vs self-managed, cost comparison, operational complexity, and enterprise adoption trends"
## ❌ Bad Plan (What NOT to Do)
### Phase 1: Assessment (Insufficient)
**Vague Assessment**:
- Need to research Kubernetes
- Compare different approaches
- Look at costs
- Check what people are using
**Problems**:
- No specific questions identified
- No data points enumerated
- No output format defined
- No user priorities understood
---
### Phase 2: Query Classification (Skipped)
**No Classification Performed**:
Just jumped straight to deploying subagents without thinking about query type.
**Problems**:
- Missed opportunity to select optimal strategy
- No rationale for subagent count
- Random delegation without systematic approach
- Reduces research quality and efficiency
---
### Phase 3: Research Plan (Poorly Structured)
#### Vague Subagent Allocation: 10 Subagents (Too Many, Too Granular)
**Subagent 1**: Research EKS
**Subagent 2**: Research GKE
**Subagent 3**: Research AKS
**Subagent 4**: Research kubeadm
**Subagent 5**: Research kops
**Subagent 6**: Research costs
**Subagent 7**: Research operational stuff
**Subagent 8**: Research what companies use
**Subagent 9**: Research Kubernetes in general
**Subagent 10**: Compare everything
**Problems with This Allocation**:
1. **Overly Granular** (Anti-Pattern):
- 10 subagents for medium-complexity query is excessive
- Should be 3-5 subagents maximum
- Subagents 1-3 could be combined: "Hosted Kubernetes landscape"
- Subagents 4-5 could be combined: "Self-managed Kubernetes landscape"
- Creates unnecessary coordination overhead
2. **Vague Instructions**:
- "Research EKS" - research what about it? Features? Pricing? Use cases?
- "Research costs" - costs of what? How compare? What scenarios?
- "Research operational stuff" - what operations? Complexity? Time? Tools?
- No specific objectives, sources, or data points
3. **Unclear Scope Boundaries**:
- Subagent 6 "costs" overlaps with Subagents 1-5 (each service has costs)
- Subagent 9 "Kubernetes in general" overlaps with everything
- Subagent 10 "compare everything" overlaps with all others
- High likelihood of duplicated effort
4. **No Aggregation Strategy**:
- How will 10 separate reports be combined?
- What structure will organize findings?
- No synthesis plan defined
---
### Example of Bad Subagent Instructions
**Subagent 6: Research costs**
"Research Kubernetes costs. Look at how much it costs to run Kubernetes. Find pricing information. Write a report about costs."
**Why This Is Bad**:
1. **No Specific Objectives**:
- "Research costs" is too vague
- Costs of what exactly? Infrastructure? Operations? Tooling?
- For what scenarios? Small clusters? Large? Which providers?
2. **No Source Guidance**:
- Where should research begin?
- What types of sources are authoritative?
- Primary vs secondary sources?
- No prioritization
3. **No Data Points Specified**:
- What specific numbers needed?
- What granularity? (per node? per cluster? per hour?)
- What cost components? (compute, storage, network, labor?)
4. **No Output Format**:
- "Write a report" is not specific
- What structure? What density?
- Comparison table? Narrative? Scenarios?
5. **No Scope Boundaries**:
- Should this cover hosted AND self-managed costs?
- Does it overlap with other subagents researching specific providers?
- No clarity on what NOT to research
6. **No Quality Criteria**:
- How recent must data be?
- What level of verification required?
- When is research complete?
---
### Phase 4: Execution (No Plan)
**Unplanned Deployment**:
- Deploy all 10 subagents at once without thought
- No monitoring plan
- No adaptation strategy
- No diminishing returns detection
**Problems**:
- Inefficient parallel execution of redundant work
- No Bayesian adaptation based on findings
- Likely to continue past diminishing returns
- May miss critical information due to poor coordination
---
## Why This Is a Bad Research Plan
### Critical Failures
#### 1. Insufficient Assessment
- Jumped to execution without understanding query
- No enumeration of key questions
- No identification of required data points
- Output format undefined
#### 2. Skipped Query Classification
- No systematic thinking about research strategy
- Random subagent allocation
- No optimization for query type
- Missed opportunity for parallel efficiency
#### 3. Overly Granular Decomposition
- 10 subagents when 3-5 optimal
- Unnecessary coordination overhead
- Violates "fewer capable subagents" principle
- Wastes resources
#### 4. Vague Subagent Instructions
- No specific objectives
- No source guidance
- No data points enumerated
- No output format
- No scope boundaries
#### 5. Unclear Scope Boundaries
- Significant overlap between subagents
- High likelihood of duplicated research
- Inefficient use of parallel execution
- Difficult synthesis due to redundancy
#### 6. No Aggregation Strategy
- Unclear how 10 reports will be combined
- No structure pre-planned
- Synthesis ad-hoc rather than systematic
#### 7. No Execution Plan
- No monitoring for progress
- No adaptation strategy
- No diminishing returns detection
- Inefficient and likely incomplete
### Likely Outcomes
Following this bad plan would produce:
❌ **Inefficient Research**:
- Wasted effort on duplicated work
- 10 subagents when 4 would suffice
- Coordination overhead exceeds value
❌ **Incomplete Coverage**:
- Vague instructions likely miss critical data
- No verification of key questions answered
- Gaps in research due to poor planning
❌ **Poor Quality Sources**:
- No source prioritization guidance
- Subagents may use low-quality sources
- No verification protocol
❌ **Difficult Synthesis**:
- Redundant findings across 10 reports
- Unclear how to structure final output
- High likelihood of missing connections
❌ **Low Information Density**:
- Vague instructions produce vague reports
- Lack of specific data points
- More words, less insight
❌ **Delayed Delivery**:
- Excessive subagent count takes longer
- May continue past diminishing returns
- Inefficient process overall
---
## How to Fix This Bad Plan
### Apply Good Research Coordination Principles
1. **Start with Systematic Assessment**:
- List ALL key questions explicitly
- Enumerate required data points
- Define output format upfront
- Understand user priorities
2. **Always Classify Query Type**:
- Depth-first, breadth-first, or straightforward?
- Select strategy matching query type
- Plan subagent count accordingly
3. **Optimize Subagent Count**:
- Combine Subagents 1-3 → "Hosted Kubernetes Landscape"
- Combine Subagents 4-5 → "Self-Managed Kubernetes Landscape"
- Keep Subagent 6 as "Cost Comparison Analysis"
- Keep Subagent 8 as "Enterprise Adoption Trends"
- Remove Subagents 7, 9, 10 (redundant or vague)
- Result: 4 capable subagents instead of 10 narrow ones
4. **Write Extremely Detailed Instructions**:
- 1-4 specific objectives per subagent
- Sources explicitly identified
- Exact data points enumerated
- Output format defined
- Scope boundaries clear
5. **Define Clear Scope Boundaries**:
- Each subagent knows exactly what to cover
- Each knows what NOT to cover
- Minimal overlap, maximum efficiency
6. **Plan Aggregation Strategy**:
- Pre-define final report structure
- Identify cross-stream dependencies
- Document synthesis approach
7. **Plan Execution Carefully**:
- Parallel deployment where appropriate
- Monitoring for progress and gaps
- Bayesian adaptation based on findings
- Diminishing returns detection
### Result: Good Research Plan
Following these fixes transforms the bad plan into a systematic, efficient, high-quality research coordination effort that produces actionable, comprehensive results.
See `good-research-plan.md` for the properly structured version.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |