ValorInvestigator/deep-research
skills/deep-research/SKILL.md
Autonomous deep OSINT research engine for the Bingaman/McSherry guardianship case. Builds intelligence reports through iterative "waves" of web searches, Apify scraping, cross-referencing, and lead generation. Each wave produces new findings that generate leads for the next wave. Uses 132+ Apify actors across 9 categories, 7 OSINT MCP tools (Sherlock, Maigret, Holehe, TheHarvester, SpiderFoot, GHunt, Blackbird), CourtListener API, Firecrawl, and the full osint-skill social media arsenal (55+ actors for Instagram, Facebook, TikTok, YouTube, Google Maps). ALWAYS trigger this skill when Levi says ANY of these (including typos): - "deep research" / "deep reserach" / "deep reseach" - "continue deep research" / "start deep research" / "resume research" - "keep researching" / "run more waves" / "next wave" - "continue OSINT" / "continue scorched earth OSINT" / "continue operation scorched earth" - "scrape" / "investigate" / "dig into" - "search for" / "search the web" / "look on the web" / "look online" - "find out about" / "find info on" / "find information" - "look into" / "look up" / "lookup" - "research" / "reserach" / "reasearch" / "reseach" - "what can you find on" / "what do we know about" - "who is" (followed by a person name in the investigation) - "check on" / "pull up" / "look at" (when referring to a person, org, or entity) - "web search" / "online search" / "google" - "osint" / "intel" / "intelligence" - "apify" / "scrape" / "crawl" - "court records" / "courtlistener" / "court listener" - "deep research scrape" / "full arsenal" / "hit it with everything" - "username search" / "email search" / "sherlock" / "maigret" / "holehe" This skill is designed for COLD START recovery. Even with a brand new PC, a fresh Claude Code install, no saved sessions -- if the data files exist, this skill tells Claude exactly how to pick up where it left off.
tools
Updated Apr 22, 2026
$ install --global
skillsauthnpx skillsauth add ValorInvestigator/claude-plugin-toolkit 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, 10:56 PM47.4s16 files scanned
SKILL.md
- name:
- deep-research
- description:
- |
- user_invocable:
- true
Deep Research -- Autonomous OSINT Intelligence Engine v6.0
RULE ZERO: NEVER STOP
When Levi starts deep research, he is LEAVING. Run wave after wave autonomously until:
- Hard technical wall (API down, budget exhausted, rate limited)
- Context window about to overflow
- Levi explicitly says stop
KEEP GOING UNTIL YOU CAN'T.
CRITICAL: BANNED TOOL -- apify/rag-web-browser
NEVER use the apify-slash-rag-web-browser tool (also called mcp__Apify__apify-slash-rag-web-browser).
This tool has been REMOVED from the MCP server configuration. It is a weak, general-purpose Google search + scrape that:
- Returns bloated HTML instead of clean structured data
- Cannot handle JS-rendered sites
- Cannot do bulk extraction
- Wastes tokens with irrelevant content
- Is the WRONG tool for OSINT research
If you find yourself about to call apify-slash-rag-web-browser, STOP and do this instead:
- For web search: Use
call-actorwithapify/google-search-scraper(SERP results, 10x more queries) - For page content: Use
call-actorwithapify/website-content-crawler(clean markdown) - For JS sites: Use
call-actorwithapify/playwright-scraper(headless browser) - For PDFs: Use
call-actorwithjirimoravcik/pdf-text-extractor(free) - For news: Use
call-actorwitheasyapi/google-news-scraper(structured news) - For court records: Use
call-actorwithautomation-lab/court-records-scraper - For quick URL fetch: Use the
WebFetchbuilt-in tool (no Apify needed)
All schemas are in references/SCHEMAS.md -- no fetch-actor-details needed for these 10 actors.
ADAPTIVE RECONNAISSANCE (mandatory before scraping)
Before scraping ANY new target type, you MUST follow this 3-step pipeline:
Step 1 -- DISCOVER: Run search-actors with keywords related to the target
Example: search-actors with keywords "Oregon court records" or "nonprofit 990"
Step 2 -- VALIDATE: Run fetch-actor-details on the best match to get input schema
Example: fetch-actor-details for "automation-lab/court-records-scraper"
Step 3 -- EXECUTE: Run call-actor with the validated schema
Example: call-actor with the exact input format from Step 2
Skip Steps 1-2 ONLY for actors already in the QUICK ACTOR REFERENCE table below (schemas pre-built).
TOOL DIVERSITY ENFORCEMENT
A PreToolUse hook in settings.json HARD BLOCKS the RAG browser. Additionally:
- Run
bash ~/.claude/skills/deep-research/scripts/search-barrage.sh "query" generalto generate diverse payloads - Read
references/SCHEMAS.mdfor pre-built actor input schemas (no fetch-actor-details needed) - Use call-actor with Google SERP FIRST, then specialized actors for deep dives
SCRIPTS (run via Bash tool)
| Script | Usage | Purpose |
|--------|-------|---------|
| scripts/diagnose.sh | Run FIRST every session | Check all tool availability |
| scripts/search-barrage.sh "query" type | type: person/org/legal/general | Generate diverse tool call payloads |
| scripts/validate-wave.sh | Run before git commit | Pre-commit quality check |
| scripts/courtlistener.py --query "terms" --court or | Legal research | CourtListener API wrapper |
QUICK ACTOR REFERENCE
| Category | Go-To Actor | Free? |
|----------|------------|-------|
| Google SERP | apify/google-search-scraper | No |
| Web Crawling | apify/website-content-crawler | Yes |
| News | easyapi/google-news-scraper | No |
| Email/Contact | caprolok/website-email-phone-finder | No |
| Court Records | automation-lab/court-records-scraper | No |
| PDF Extraction | jirimoravcik/pdf-text-extractor | Yes |
| Headless Browser | apify/playwright-scraper | Yes |
Full catalog: references/ACTOR_INVENTORY.md | Pre-built schemas: references/SCHEMAS.md
OSINT MCP TOOLS (Docker)
| Tool | Input | What It Does |
|------|-------|-------------|
| sherlock_username_search | username | 399+ social media sites |
| maigret_username_search | username | 3000+ sites |
| holehe_email_search | email | 120+ platform check |
| theharvester_domain_search | domain | Emails, subdomains |
Full reference: references/OSINT_MCP_TOOLS.md
CONFIDENCE GRADING
- A: 2+ independent sources or official documents
- B: 1 credible source (LinkedIn, court record, news)
- C: Indirect evidence (metadata, connections, inference)
- D: Single unverified mention
COLD START RECOVERY
Read in order:
references/FILE_LOCATIONS.md-- paths, tokens, git workflowmemory/MEMORY.md-- master indexmemory/active_work.md-- wave logmemory/OPEN_LOOPS.md-- open leadsreports/NETWORK_INTEL_REPORT.md-- first 10 lines (version, wave count)
WAVE PROTOCOL (summary)
Phase 1: Search barrage (8-10 parallel, multiple tool types) Phase 2: Deep dive (2-3 best findings, specialized actors) Phase 2.5: Gap analysis (every 5th wave) Phase 3: Integrate (update report, sources, leads) Phase 4: Memory update (active_work, OPEN_LOOPS) Phase 5: Git commit and push
Full protocol: references/WAVE_PROTOCOL.md
INVESTIGATION CONTEXT
Eastern Oregon healthcare governance network. Same people control hospital boards, Medicaid CCO ($600M+), behavioral health org, and employ attorneys on both sides of guardianship disputes.
Key entities: EOCCO, GOBHI, Moda/ODS, Good Shepherd/GRH Foundation, Baum Family, OJD/Keffer
Full playbooks: references/INVESTIGATION_PLAYBOOKS.md
COURTLISTENER
Token: 48b2b5fd5858fe301286d216f6d1f4146507910c
Base URL: https://www.courtlistener.com/api/rest/v4/
Quick search: python3 scripts/courtlistener.py --query "terms" --court or
REFERENCE FILE INDEX
| File | Contents |
|------|----------|
| references/WAVE_PROTOCOL.md | Full 5-phase protocol + gap analysis |
| references/SCHEMAS.md | 10 pre-built actor input schemas |
| references/INVESTIGATION_PLAYBOOKS.md | Person/org/document investigation recipes |
| references/OSINT_MCP_TOOLS.md | 7 Docker OSINT tools |
| references/SOCIAL_MEDIA_ACTORS.md | 55+ social media actors |
| references/SEARCH_HIERARCHY.md | 8-tier search engine priority |
| references/ACTOR_INVENTORY.md | 132-actor catalog |
| references/FILE_LOCATIONS.md | Paths, tokens, git workflow |
CURRENT STATUS (v6.0, March 21, 2026)
- Report: 950+ leads, 155 waves, NETWORK_INTEL_REPORT.md v98.0
- New: PreToolUse hook enforces tool diversity, embedded scripts, subagent definition
- Apify budget: $20/month. BE AGGRESSIVE.
VERSION HISTORY
- v1.0 (Mar 8): Initial skill, 60 waves
- v2.0 (Mar 14): Added full Apify arsenal
- v3.0 (Mar 16): Added skip trace, Firecrawl
- v4.0 (Mar 17): Complete rebuild, 132 actors, progressive disclosure
- v5.0 (Mar 21): Added 7 OSINT MCP tools (Docker), 55+ social media actors, confidence grading
- v5.1 (Mar 21): Added Phase 2.5 Gap Analysis
- v5.2 (Mar 21): Added pre-built schemas, hard enforcement rule, minimum wave composition
- v6.0 (Mar 21): Full architectural rebuild. PreToolUse hook enforcement, embedded bash scripts (diagnose, search-barrage, validate-wave, courtlistener), specialized deep-researcher subagent, progressive disclosure (491 -> 180 lines). Governance layer: hooks enforce, skills instruct, agents execute.
Related Skills
ValorInvestigator/skills/write-article
development
VerifiedTrustedCommunity
# Write Article -- Investigative Series in Levi Bakke's Voice You are ghostwriting publishable investigative journalism in Levi's voice. He is a participant-investigator -- IN the story, not observing from outside. ## BEFORE WRITING Read the style guide: [references/style-guide.md](references/style-guide.md) Read the gold standard: `C:\Users\Big Levi\Desktop\DHS Stories\the Canary FINAL.txt` ## THE WRITING PROCESS 1. **Gather** -- Read relevant timeline docs, investigation files, databases
SKILL.mdUpdated Apr 22, 2026
ValorInvestigator/web-search
development
VerifiedTrustedCommunity
Dual-engine web search using BOTH Firecrawl AND Brave Search simultaneously. ALWAYS trigger this skill when Levi uses any of these phrases or close variations: - "search the web" / "search the internet" / "search online" - "www" (used as a verb or shorthand, e.g. "www this", "look it up on the www") - "internet" (as in "check the internet", "find on the internet", "look this up on the internet") - "go online", "look this up online", "check online" - "search for X" when context implies web search (not local files or database) - "find X online", "look up X", "research X on the web" This is Levi's preferred web research protocol. Both engines run together -- Brave for fast broad coverage, Firecrawl for deep scraping. Never use just one without the other when this skill triggers.
SKILL.mdUpdated Apr 22, 2026
ValorInvestigator/web-scraping
development
VerifiedTrustedCommunity
Web scraping with anti-bot bypass, content extraction, undocumented APIs and poison pill detection. Use when extracting content from websites, handling paywalls, implementing scraping cascades or processing social media. Covers requests, trafilatura, Playwright with stealth mode, yt-dlp and instaloader patterns.
SKILL.mdUpdated Apr 22, 2026
ValorInvestigator/skills/text-to-voice
development
VerifiedTrustedCommunity
# Text to Voice -- Convert Articles to Audio Convert written articles to spoken audio (.mp3) using Google Cloud TTS with Chirp 3: HD Algieba voice. ## VOICE PROFILE - **Voice:** `en-US-Chirp3-HD-Algieba` (male, Chirp 3: HD) - **Speaking Rate:** `1.0` | **Volume Gain:** `0.0` dB - **Audio Encoding:** MP3, 44100 Hz, 192k bitrate (final stitch) - **API Version:** `texttospeech_v1beta1` (Chirp 3 HD requires v1beta1) - **Google Cloud Project:** `valorinvestigates` ## THE TWO-STEP PROCESS 1. **Rew
SKILL.mdUpdated Apr 22, 2026