Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
samhvw8 avatar

Lead Researcher

  • 4 installs
  • 14 repo stars
  • Updated July 28, 2026
  • samhvw8/dotfiles

Helps with ai & agent building tasks.

About

lead-researcher is a Claude Code skill for ai & agent building. It helps solo builders move faster with AI-assisted development.

  • lead-researcher
  • AI & Agent Building
  • AI-coding skill

Lead Researcher by the numbers

  • 4 all-time installs (skills.sh)
  • Ranked #13,349 of 16,556 AI & Agent Building skills by installs in the Skillselion catalog
  • Data as of Jul 29, 2026 (Skillselion catalog sync)
npx skills add https://github.com/samhvw8/dotfiles --skill lead-researcher

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Installs4
repo stars14
Last updatedJuly 28, 2026
Repositorysamhvw8/dotfiles

What it does

Helps with ai & agent building tasks.

Files

SKILL.mdMarkdownGitHub ↗

Lead Researcher

Orchestrator-worker pattern for complex research. You are the lead agent — you plan, decompose, delegate to parallel sub-agents, synthesize, and iterate.

Based on: https://www.anthropic.com/engineering/multi-agent-research-system

When This Skill Applies (Complexity Test)

Use lead-researcher when 2+ of these are true:

SignalExample
Multi-domain"Compare auth approaches across mobile, web, and IoT"
Multi-perspective"Evaluate X from security, performance, and DX angles"
Contradictory landscapeTopic where communities disagree
Breadth + depth needed5+ independent sub-questions
Multi-language valueChinese/Russian communities have distinct signal
Synthesis-heavyAnswer requires weighing tradeoffs across findings

If only 0-1 signals match → use research skill directly (single-agent research).

Workflow

Phase 1: Plan (You, the Lead)

Before spawning agents, think through:

1. Decompose the question into 3-5 independent sub-questions 2. Assess complexity per sub-question (simple=1 agent, comparison=2, deep=3) 3. Assign each sub-agent: objective, output format, source guidance, boundaries 4. Budget total effort: max 3 parallel agents per wave

Write a brief plan (3-5 bullets) before delegating. Save it — context will grow large.

Failure mode to avoid: Spawning agents without clear task boundaries → duplicated work, gaps, wasted tokens.

Phase 2: Delegate (Parallel Sub-Agents)

Spawn sub-agents with the researcher agent type. Each gets:

[Specific objective — what to find]
[Output format — what to return]
[Source guidance — where to look, what languages]
[Boundaries — what NOT to investigate, what's handled by other agents]

RECOMMENDED SKILLS: research - use for search methodology and gh CLI patterns

Scaling rules:

ComplexityAgentsTool calls each
Fact-finding13-10
Comparison2-310-15
Deep multi-field3 (max per wave)15-30

Fire all independent agents in a single message (parallel execution).

Phase 3: Synthesize (You, the Lead)

When agents return:

1. Evaluate — Are there gaps? Contradictions? Missing angles? 2. Cross-reference — Do findings from different agents agree? Where do they conflict? 3. Decide — Is another wave needed? (max 2 waves total, 6 agents total) 4. Merge — Combine into coherent findings with source attribution

Iteration trigger: Only spawn wave 2 if wave 1 revealed a critical gap you couldn't have predicted upfront.

Phase 4: Report

Structure the final synthesis:

  • Lead with the answer/recommendation
  • Present tensions between sources honestly
  • Every claim MUST cite its source — inline [text](url) next to each claim
  • List unresolved questions at end
  • End with ## Sources section listing all URLs used
  • No claim without a source — if you can't cite it, don't include it

Anti-Patterns

TrapFix
5+ agents for simple queryCheck complexity test — probably just research skill
All agents searching same thingGive explicit non-overlapping boundaries
Endless iteration wavesMax 2 waves. Ship what you have.
Over-investing in one angleBalance effort across sub-questions
No plan before spawningMUST write decomposition first

Quality Check

Before reporting to user:

  • [ ] All claims have a source
  • [ ] Conflicting findings are presented as tensions, not hidden
  • [ ] Each sub-question got adequate coverage
  • [ ] Report is concise — sacrifice grammar for brevity
  • [ ] Report saved to ./research/YYMMDD-<topic>.md

Related skills

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.