nsheaps/deep-research
plugins/deep-research/skills/deep-research/SKILL.md
Use this skill when investigating complex questions that require multiple sources, angles, or synthesis. The deep-research plugin provides a 3-agent research team: lead-researcher (orchestrator), sub-researcher (workers), and critical-reviewer (validator). Trigger when: - Investigation requires 3+ sources (docs, issues, source code, web) - You need to synthesize findings across multiple sources into a coherent report - Previous research findings need source-level verification - You're investigating behavior differences between documented and observed behavior - Competitive analysis or technology evaluations Do NOT trigger for: - Simple lookups ("what flag does X?") - Single-source answers - Basic codebase navigation (use Grep/Glob directly)
$ install --global
skillsauthnpx skillsauth add nsheaps/ai-mktpl deep-researchInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
Security Scan Results
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
3
Scanners Passed
9
Scanners in report
Clean
TrivyContainer and dependency vulnerability scanner
95%
Clean
SemgrepStatic code analysis for vulnerabilities
95%
Clean
mcp-scan (Snyk)Model Context Protocol security validation
95%
Skipped
Snyk (dep)Open source security scanning
50%
Skipped
Socket.devSupply chain security analysis
50%
Skipped
VirusTotalMulti-engine malware detection
50%
Skipped
CrowdStrikeAdvanced threat intelligence
50%
Skipped
OSV-ScannerOpen Source Vulnerability database check
50%
Skipped
OWASP Dep-Check
50%
Last scanned: Apr 22, 2026, 4:08 PM26.4s1 file scanned
SKILL.md
- name:
- deep-research
- description:
- |
- Use this skill when investigating complex questions that require multiple sources, angles, or synthesis. The deep-research plugin provides a 3-agent research team:
- lead-researcher (orchestrator), sub-researcher (workers), and critical-reviewer (validator).
Deep Research -- 3-Agent Research System
A multi-agent research system for complex, evidence-based investigations.
Architecture
The deep-research plugin uses three specialized agents:
User Request
|
v
lead-researcher (orchestrator)
|-- Plans 5-10 search angles
|-- Dispatches sub-researchers (one per angle)
| |-- sub-researcher 1: Angle A (web + docs)
| |-- sub-researcher 2: Angle B (source code)
| |-- sub-researcher 3: Angle C (community)
| |-- ... (up to 10 angles)
|
|-- Collects findings from all sub-researchers
|-- Dispatches critical-reviewer to validate
|-- Addresses gaps from review
|-- Synthesizes final report
|
v
Report saved to .claude/tmp/research-<topic>.md
Agents
| Agent | Role | Tools | Restrictions | | --------------------- | ------------------------------------------------------------------------ | -------------------------------------------- | -------------------------------------------- | | lead-researcher | Orchestrator -- plans angles, dispatches workers, synthesizes report | Read, Write, Grep, Glob, Agent | No web search (delegates to sub-researchers) | | sub-researcher | Worker -- investigates one specific angle, writes findings to file | Read, Write, Grep, Glob, WebSearch, WebFetch | No Edit (shouldn't modify existing files) | | critical-reviewer | Validator -- challenges assumptions, identifies gaps, suggests follow-up | Read, Grep, Glob, WebSearch, WebFetch | No Edit, no Write (read-only reviewer) |
How to Invoke
For complex, multi-source investigations, spawn the lead-researcher agent:
Agent(lead-researcher, "Investigate how Claude Code spawns teammates -- check source code, official docs, GitHub issues, and community implementations. Save report to .claude/tmp/research-teammate-spawning.md")
The lead-researcher will autonomously:
- Plan research angles
- Dispatch sub-researchers for each angle
- Collect and cross-reference findings
- Run critical review
- Produce final report with confidence levels and citations
When to Use Deep Research vs Simple Lookups
| Situation | Use | | ------------------------------------------------ | --------------------------------------------------------- | | "What flag enables X?" | Direct answer -- no agent needed | | "How does X work internally?" | lead-researcher -- multi-source investigation | | "Compare X vs Y vs Z for our use case" | lead-researcher -- parallel sub-researchers per option | | "Find the file that does X" | Grep/Glob -- no agent needed | | "Why does X behave differently than documented?" | lead-researcher -- needs source + docs + issues | | "What's the best practice for X?" | Context7 or WebSearch -- single source usually sufficient |
Output Format
Reports are saved to .claude/tmp/research-<topic>.md with:
- Executive Summary: 2-3 sentence answer
- Methodology: Angles pursued, sources consulted
- Findings: Each finding with confidence level and citations
- Critical Review Notes: What was challenged, how addressed
- Open Questions: What remains unanswered
- Sources: Complete list of all references
Confidence Level Framework
| Level | Meaning | Evidence Required | | :-------------- | :------------------------------------------- | :-------------------------------------- | | Very High | Confirmed from source code or official specs | Primary source + verified | | High | Multiple independent sources agree | 3+ sources or official docs + community | | Medium-High | Strong evidence with some inference | 2 sources + logical reasoning | | Medium | Plausible with supporting evidence | 1 source + consistent behavior | | Low | Hypothesis based on limited evidence | Inference only |
Tips
- Save everything to files -- Context compaction destroys in-memory state. Files survive.
- The lead-researcher dispatches sub-researchers -- You don't need to manage them manually.
- Critical review catches gaps -- The critical-reviewer challenges weak evidence before the report is finalized.
- Confidence levels are mandatory -- Every finding needs one.
- Correct previous research openly -- Being wrong in early phases is expected and documented.
References
- Claude Code Agents documentation
- Claude Code Plugins documentation
Related Skills
nsheaps/github-app-session-env
tools
VerifiedTrustedCommunity
Manually reproduce what the github-app plugin's SessionStart hook does to make a GitHub App installation token usable in the current session — materialize the PEM, generate the token, isolate GH_CONFIG_DIR, write the runtime env file, and wire CLAUDE_ENV_FILE so every Bash call sees GH_TOKEN/GITHUB_TOKEN. Use when the hook did not run, the token is missing from the environment, or a shell/teammate needs the token wired up by hand. <example>GH_TOKEN isn't set even though github-app is configured</example> <example>the github-app SessionStart hook didn't run, set up the token manually</example> <example>wire the github app token into CLAUDE_ENV_FILE</example> <example>gh keeps falling back to the wrong account, isolate GH_CONFIG_DIR</example>
3SKILL.mdUpdated Jun 9, 2026
nsheaps/github-app-git-identity
tools
VerifiedTrustedCommunity
Manually configure the GitHub App bot git identity the way the github-app plugin's SessionStart hook does — resolve the app slug and bot user ID, build the <slug>[bot] name and noreply email, set GIT_AUTHOR_*/GIT_COMMITTER_* env vars, and write an isolated GIT_CONFIG_GLOBAL with the gh auth git-credential helper. Use when commits are attributed to the wrong account, "Author identity unknown" appears, or git identity must be set up by hand. <example>my commits are showing up as the handler, not the bot</example> <example>git says Author identity unknown after the github-app hook ran</example> <example>configure the github app bot git identity manually</example> <example>set up the gh credential helper for git push</example>
3SKILL.mdUpdated Jun 9, 2026
nsheaps/spec-management
tools
VerifiedTrustedCommunity
Manages spec files for requirements capture and validation
3SKILL.mdUpdated Jun 7, 2026
nsheaps/plugins/bash-command-rejection/skills/bash-chaining-alternatives
tools
VerifiedTrustedCommunity
# Bash Chaining Alternatives This skill teaches you how to work around the bash command chaining restriction enforced by this plugin. ## Why Chaining is Blocked The `bash-command-rejection` plugin blocks these operators: | Operator | Name | Why Blocked | | -------- | ---------- | ----------------------------------------------------------------------------------- | | `&&` | AND chain | Runs cmd2 only if cmd1 su
3SKILL.mdUpdated Jun 7, 2026