SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillscti-domain-research

cti-domain-research

Security & auditing

View repository

Updated 10-16 02:42

About this skill

CTI Domain Research

Structured threat intelligence search across 300+ curated security domains with optional NotebookLM ingestion.

Customization

Configure these settings to personalize your CTI research workflow. All are optional — the skill works with sensible defaults.

Setting Description Default How to Set
Search provider Which search API to use Brave Search SEARCH_PROVIDER=brave in .env
Brave API key Your Brave Search API key — (required) BRAVE_SEARCH_API_KEY=... in .env
SerpAPI key Alternative search provider — SERPAPI_KEY=... in .env
Google CSE key + ID Alternative search provider — GOOGLE_CSE_KEY=... + GOOGLE_CSE_ID=... in .env
NotebookLM notebook ID Target notebook for --notebooklm flag — NOTEBOOKLM_NOTEBOOK_ID=... in .env or --notebook-id flag
Default result count Results per tier 10 --count N flag per query
Default recency How far back to search 90 days --since DAYS flag per query
Custom domains Add your own security sources 595 built-in Edit data/domains.txt and data/tier-map.json

Setup steps:

  1. Copy .env.example to .env in the plugin directory
  2. Add your search API key (Brave recommended — 2,000 free requests/month)
  3. Optionally set your NotebookLM notebook ID
  4. Restart Claude Code
cd ${CLAUDE_PLUGIN_ROOT}
cp .env.example .env
# Edit .env with your API keys

Adding custom domains:

# Add a domain (one per line)
echo "your-security-blog.com" >> data/domains.txt

# Add tier mapping
# Edit data/tier-map.json and add:
# "your-security-blog.com": { "tier": 3, "authority": 70 }

Workflow

1. Parse query + flags
2. Select domain tier(s) based on query type
3. Execute scoped web searches (site: operator per domain batch)
4. Deduplicate + rank results by recency and source authority
5. Extract: title, source, date, summary, key IOCs/TTPs/CVEs mentioned
6. [Optional] Push findings to NotebookLM via connector plugin
7. Return structured CTI brief

Query Parsing

Accept free-text queries. Extract:

  • Subject: CVE ID, actor name, malware family, technique (MITRE ATT&CK), product name
  • Time filter: default = last 90 days; honour "last week / month / year" if stated
  • Depth: default = top 10 results; --count N overrides
  • Output format: default = brief table + bullets; --full = long-form brief; --json = raw JSON
  • NotebookLM push: --notebooklm flag triggers the connector (see NotebookLM section below)

Domain Tiers

Load the full domain list from references/domains.txt. Apply tier routing:

Tier When to use Examples
Tier 1 — Authoritative CVEs, advisories, govt alerts cisa.gov, nvd.nist.gov, msrc.microsoft.com, access.redhat.com, support.apple.com, cert.europa.eu, ncsc.gov.uk
Tier 2 — Vendor Research Deep technical analysis unit42.paloaltonetworks.com, blog.talosintelligence.com, securelist.com, research.checkpoint.com, thedfirreport.com, blog.google, mandiant.com, secureworks.com, sentinelone.com, crowdstrike.com
Tier 3 — News + Community Situational awareness bleepingcomputer.com, krebsonsecurity.com, therecord.media, thehackernews.com, arstechnica.com, reddit.com/r/cybersecurity, news.ycombinator.com
Tier 4 — Specialised Malware, PoC, OSINT any.run, otx.alienvault.com, vulncheck.com, attackerkb.com, greynoise.io, hunt.io, packetstormsecurity.com

Routing logic:

  • CVE query → Tier 1 first, then Tier 2
  • Threat actor / malware → Tier 2 first, then Tier 4
  • Situational / news → Tier 3 first, then Tier 2
  • PoC / exploit → Tier 4 + Tier 2
  • General → all tiers, ranked by authority score

Search Execution

Use the web search tool with site: scoping. Batch domains to maximise coverage per call:

# Example batch query construction
query = f'"{subject}" site:bleepingcomputer.com OR site:krebsonsecurity.com OR site:therecord.media'

# For CVEs, always include NVD and MSRC
query = f'{cve_id} site:nvd.nist.gov OR site:msrc.microsoft.com OR site:cisa.gov'

Run multiple searches in parallel when --full is specified:

  1. Tier 1 batch
  2. Tier 2 batch
  3. Tier 3 + 4 batch

Result Schema

For each result, extract:

{
  "title": "...",
  "source": "domain.com",
  "url": "https://...",
  "published": "YYYY-MM-DD",
  "authority_tier": 1,
  "summary": "1-2 sentence distillation",
  "tags": ["CVE-XXXX", "ransomware", "RCE", "MITRE:T1190"],
  "iocs": ["hash", "IP", "domain"],
  "severity": "Critical|High|Medium|Low|Info"
}

Output Formats

Default — CTI Brief (table + bullets)

## CTI Brief: <query> — <date>

### Key Findings
- [Critical] <1-line finding> — <source> (<date>)
- [High] ...

### Source Table
| Source | Title | Date | Tags |
|--------|-------|------|------|
| bleepingcomputer.com | ... | ... | ransomware, RCE |

### IOC Summary
IPs: ...  Hashes: ...  Domains: ...

### Next Steps
- Patch priority: ...
- Detection: MITRE T-IDs to hunt
- Intel gaps: ...

--full — Long-Form Brief

Add: executive summary, per-source analysis paragraphs, MITRE ATT&CK mapping table, recommended detection rules (Sigma header stubs).

--json

Return raw JSON array of result objects (schema above).

NotebookLM Integration

When --notebooklm flag is set, use the NotebookLM connector plugin to ingest findings.

Plugin location (Claude Code): ~/.claude/plugins/notebooklm-connector/
Plugin repo: https://github.com/Security-Phoenix-demo/security-skills-claude-code/tree/main/plugins/notebooklm-connector

Connector flow:

1. Collect all result URLs from the search phase
2. Format as source list per plugin spec
3. Call connector to add sources to target NotebookLM notebook
4. Report: N sources added, notebook URL

Plugin call pattern (Claude Code):

# Check connector is installed
ls ~/.claude/plugins/notebooklm-connector/

# Call connector with source list
node ~/.claude/plugins/notebooklm-connector/index.js \
  --notebook-id "$NOTEBOOKLM_NOTEBOOK_ID" \
  --sources "$(echo $URLS | jq -R -s 'split("\n")')" \
  --title "CTI: $QUERY — $(date +%Y-%m-%d)"

Required env var:

export NOTEBOOKLM_NOTEBOOK_ID="<your-notebook-id>"
# Or pass via --notebook-id flag at query time

Fallback (Claude.ai — no plugin runtime):

If running in Claude.ai (no bash), present a formatted source list the user can manually add to NotebookLM, plus a direct link to their notebook if the ID is known.

Slash Command (Claude Code)

Install by creating ~/.claude/commands/cti-search.md:

---
description: "Search 300+ security domains for CTI on any topic, CVE, actor, or malware"
argument-hint: "<query> [--count N] [--full] [--json] [--notebooklm] [--notebook-id ID]"
allowed-tools: WebSearch, Bash
---

Search curated security domains for threat intelligence on: $ARGUMENTS

Follow the CTI Domain Research skill (cti-domain-research) exactly.
Domain list: ~/.claude/commands/cti-domains.txt

Flags:
- --count N        : Return N results (default 10)
- --full           : Long-form brief with MITRE mapping
- --json           : Raw JSON output
- --notebooklm     : Push sources to NotebookLM after search
- --notebook-id ID : Target notebook ID (overrides env var)
- --tier 1|2|3|4   : Restrict to specific domain tier
- --since DAYS     : Limit to last N days (default 90)

Installation Script

Run scripts/install.sh to:

  1. Copy slash command to ~/.claude/commands/
  2. Copy domain list to ~/.claude/commands/cti-domains.txt
  3. Verify NotebookLM connector plugin presence
  4. Print setup confirmation

See scripts/install.sh for full details.

Error Handling

Condition Behaviour
No results for domain batch Skip tier, try next; warn if all tiers empty
NotebookLM connector not installed Warn + fallback to manual source list
Rate limited on search Back off 2s, retry once; surface partial results
Query too broad (>1000 potential results) Auto-narrow to Tier 1+2 only, suggest refinement
CVE not yet in NVD Note "CVE may be pre-NVD — check vendor advisories directly"

Reference Files

  • references/domains.txt — Full curated domain list (300+ sources, one per line)
  • references/tier-map.json — Domain → tier + authority score mapping
  • scripts/install.sh — One-shot installer for Claude Code slash command