| name | patent-intelligence-engine |
| description | Domain-specialized research engine for intellectual property and patent landscape analysis. Multi-agent pipeline with tiered depth, iterative refinement, and structured confidence scoring for IP attorneys, technology transfer officers, and R&D strategists. |
Patent Intelligence Engine Research Engine
Launch a comprehensive multi-agent research system specialized for Intellectual property and patent landscape analysis research, serving IP attorneys, technology transfer officers, and R&D strategists.
This engine implements a seven-phase research pipeline with tier-based depth routing. It is fully self-contained -- no external research plugin dependencies are required. Agent protocols, quality standards, and detailed instructions are in reference files that agents load on-demand.
Usage
/research [topic] -- Standard tier (default): 2 agents: Patent Search Specialist, Prior Art Analyst
/research [URL] -- Research starting from a specific webpage (Standard tier)
/research [topic] --quick -- Quick tier: Single-agent lookup using Patent Search Specialist
/research [topic] --deep -- Deep tier: Full pipeline with 3 agents: Patent Search Specialist, Prior Art Analyst, IP Landscape Mapper
/research [topic] --comprehensive -- Comprehensive tier: All 3 agents + follow-up round
/research [topic] --outline-only -- Planning phase only (produces outline, then stops)
/research [topic] --approve -- Pause for user approval after Phase 1 (default for Standard+ tiers)
/research [topic] --no-approve -- Skip outline approval gate (for automation or fast iteration)
/research [topic] --extend -- Build on prior research in this project (default: standalone)
/research [topic] --reverifiable -- Store source snapshots alongside hashes for independent re-verification
/research [topic] --no-vvc -- Skip VVC verification phases (Phases 5-6)
Research Architecture
| Phase | Name | Description |
|---|
| 0 | Tier Detection | Parse flags, configure paths and depth |
| 1 | Research Planning | Strategic framework development and agent task design |
| 2 | Parallel Research | Multiple specialist agents research simultaneously |
| 3 | Synthesis | Multi-source integration, contradiction resolution, gap analysis |
| 4 | Draft Reporting | Draft report generation with claim tagging for VVC |
| 5 | VVC-Verify | Verify draft claims against sources (see vvc-pipeline.md) |
| 6 | VVC-Correct | Apply corrections, produce final report (see vvc-pipeline.md) |
| Tier | Planning | Research Agents | Synthesis | Report | VVC | Provenance | User Gate |
|---|
| Quick | No | patent-intelligence-engine:patent-search-specialist | No | Inline | None | Hash-only | No |
| Standard | Yes | patent-intelligence-engine:patent-search-specialist, patent-intelligence-engine:prior-art-analyst | Yes | Draft | Verify-only | Hash + Audit | --approve only |
| Deep | Yes | patent-intelligence-engine:patent-search-specialist, patent-intelligence-engine:prior-art-analyst, patent-intelligence-engine:ip-landscape-mapper | Yes | Draft | Full | Hash + Audit | --approve only |
| Comprehensive | Yes | All 3 agents + follow-up round | Yes | Draft | Full | Hash + Audit | Always |
Phase 0: Tier Detection & Global Configuration
Parse tier from $ARGUMENTS: --quick / --standard / --deep / --comprehensive (default: Standard). Additional flags: --outline-only (stop after Phase 1), --approve / --no-approve (user gate control), --extend (project-aware context), --reverifiable (retain source snapshots), --no-vvc (skip Phases 5-6, Phase 4 becomes final reporting).
- If
--no-vvc is present, set NO_VVC to true (skip Phases 5-6, Phase 4 becomes final reporting)
- If topic starts with
http:// or https://, treat as URL seed — fetch, extract topic, derive slug from title
Derive Configuration
TOPIC_SLUG = lowercase topic, hyphenate spaces, strip punctuation
RUN_TS = YYYY-MM-DD_HHMMSS_ET (Eastern Time via bash: TZ='America/New_York' date '+%Y-%m-%d_%H%M%S_ET')
BASE_DIR = "./research-reports/${RUN_TS}_${TOPIC_SLUG}"
ENGINE_ID = "patent-intelligence-engine"
Status Directory & Pipeline State
Create BASE_DIR/_status/ directory. Write BASE_DIR/_status/_pipeline.json with initial pipeline state. Substitute all runtime variables (ENGINE_ID, TOPIC_SLUG, detected tier, agent IDs for this tier):
{
"engineId": "${ENGINE_ID}",
"topic": "${TOPIC_SLUG}",
"tier": "${DETECTED_TIER}",
"currentPhase": 0,
"phaseLabel": "Tier Detection",
"phases": [],
"agents": [],
"dashboardUrl": null,
"startedAt": "ISO-8601 timestamp",
"lastUpdated": "ISO-8601 timestamp"
}
Phases array -- build at Phase 0 based on detected tier and VVC config. Each phase object includes:
{
"phase": 0,
"label": "Tier Detection",
"status": "complete",
"startedAt": "ISO-8601 timestamp",
"completedAt": "ISO-8601 timestamp"
}
| Phase | Label | Include When |
|---|
| 0 | Tier Detection | Always |
| 1 | Research Planning | Always |
| 2 | Parallel Research | Always |
| 2.5 | Batch Source Hashing | Always |
| 3 | Research Synthesis | Always |
| 3.5 | Comprehensive Follow-Up | Comprehensive tier only |
| 4 | If --no-vvc or VVC disabled: "Professional Reporting"; otherwise: "Draft Reporting" | Always |
| 4.5 | Provenance Audit | Standard, Deep, Comprehensive |
| 5 | VVC-Verify | VVC enabled AND --no-vvc not set |
| 6 | VVC-Correct | VVC enabled AND --no-vvc not set |
All phases start with status "pending", startedAt: null, and completedAt: null -- except Phase 0 which is "complete" with both startedAt and completedAt set to the init timestamp. The agents array lists only Phase 2 research agent IDs for the detected tier.
Phase Transition Protocol
Before each phase begins, update _pipeline.json:
- Set
currentPhase and phaseLabel to the new phase
- Set the phase's status to
"in_progress"
- Set the phase's
startedAt to the current ISO-8601 timestamp
- Update
lastUpdated timestamp
After each phase completes:
- Set the phase's status to
"complete"
- Set the phase's
completedAt to the current ISO-8601 timestamp
- Update
lastUpdated timestamp
Research Dashboard (Standard, Deep, Comprehensive tiers only)
Skip this step for Quick tier.
- Copy
${CLAUDE_SKILL_DIR}/dashboard-server.js and ${CLAUDE_SKILL_DIR}/dashboard.html to BASE_DIR/_status/
- Run
node BASE_DIR/_status/server.js in background via Bash tool
- Capture the dashboard URL from stdout
- Update
_pipeline.json with dashboardUrl
- Print to user: "Research dashboard: http://localhost:"
All files live under BASE_DIR. Never restate the prompt in chat; write long data to files and keep chat responses concise.
Sub-Agent System
- research-planning-specialist
- synthesis-specialist
- research-reporting-specialist
- patent-intelligence-engine:patent-search-specialist (Patent Search Specialist)
- patent-intelligence-engine:prior-art-analyst (Prior Art Analyst)
- patent-intelligence-engine:ip-landscape-mapper (IP Landscape Mapper)
- vvc-specialist (Verification, Validation & Correction Specialist)
Research agents MUST be referenced by fully qualified name (patent-intelligence-engine:[agentId]) when deploying via Task tool.
The vvc-specialist is a pipeline agent that runs in Phases 5-6 (post-reporting). It does NOT participate in Phase 2 research.
Quick Tier
If --quick detected, deploy a SINGLE agent (patent-intelligence-engine:patent-search-specialist). Agent FIRST ACTION: Read ${CLAUDE_SKILL_DIR}/standards.md. Conduct focused patent intelligence research. Cap total WebFetch calls at 10. Format: ## Summary | ## Key Findings (with confidence) | ## Sources | ## Limitations. Skip all remaining phases.
Standard / Deep / Comprehensive Tiers
Phase 1: Strategic Research Planning
Deploy research-planning-specialist: Analyze topic complexity within IP and patent landscape analysis. Build scope grid: core questions, sub-questions, dissenting angles, geographic/temporal slices. Design task assignments for downstream agents to reduce overlap. Save outline to BASE_DIR/[TOPIC_SLUG]_Research_Outline.md.
User Gates: --outline-only: stop here | --no-approve: skip gate | Otherwise: present outline for approval.
Phase 2: Parallel Research Execution
Deploy research agents per tier config. Each agent operates independently. Each agent writes to its own files only.
Agent: Patent Search Specialist
Deploy patent-intelligence-engine:patent-search-specialist — comprehensive patent searches across USPTO, EPO, WIPO, JPO, KIPO, CNIPA, CIPO using CPC/IPC codes, keyword strategies, and assignee tracking. Documents patent numbers, dates, assignees, claim counts, classifications, and citation metrics.
Agent: Prior Art Analyst
Deploy patent-intelligence-engine:prior-art-analyst — analyzes claims for novelty/non-obviousness using TSM and KSR frameworks. Reviews prosecution histories, identifies prior art (patents, NPL, commercial products), and maps claim-by-claim relationships.
Agent: IP Landscape Mapper
Deploy patent-intelligence-engine:ip-landscape-mapper — filing trends, top assignees, geographic patterns, CPC technology clusters, competitive matrices, whitespace identification, and FTO risk mapping.
Common Agent Requirements -- every Phase 2 agent MUST:
- FIRST ACTION: Read
${CLAUDE_SKILL_DIR}/standards.md, ${CLAUDE_SKILL_DIR}/research-protocol.md, and BASE_DIR/[TOPIC_SLUG]_Research_Outline.md
- Apply Search Query Generation Protocol (4+ query types) and Iterative Search-Assess-Refine (max 4 iterations)
- Include confidence tier (HIGH/MEDIUM/LOW/SPECULATIVE) and source credibility tier (1-5) for every claim
- Save: bibliography (
_[AgentID]_Bibliography.md), claims table (_Claims_[AgentID].md), methodology log (_Methodology_Log_[AgentID].md), sources (_Sources_[AgentID].md)
- Cap total WebFetch calls at 10. If a URL returns 403/blocked/paywall, note in methodology log and move on -- do not retry.
- Run contrarian sweep: search for refutations, critiques, failure cases, contradictory data
- Recursive web exploration up to 5 levels deep from seed URLs
Phase 2 Output Verification
After all Phase 2 agents complete, verify output before proceeding to Phase 2.5:
-
For each deployed agent, check:
a. BASE_DIR/_status/[AgentID].json exists
b. Status file shows status "complete" (not "error" or "researching")
c. BASE_DIR/[TOPIC_SLUG]_Claims_[AgentID].md exists and is non-empty
d. BASE_DIR/[TOPIC_SLUG]_[AgentID]_Bibliography.md exists and is non-empty
-
Classify each agent's outcome:
- HEALTHY: status "complete" + non-empty output files
- PARTIAL: status "complete" but some output files missing/empty (possible write failure)
- FAILED: status "error" or status file missing entirely
- EMPTY: status file exists but no output files at all
-
Decision logic:
- All HEALTHY: proceed to Phase 2.5
- Mix of HEALTHY + PARTIAL/FAILED: log warnings, proceed with available data
- All FAILED/EMPTY: halt pipeline, inform user: "All Phase 2 agents failed to produce output. Check the dashboard for error details. No research data available for synthesis."
-
Log verification results to BASE_DIR/[TOPIC_SLUG]_Methodology_Log.md (this becomes the first entry in the consolidated methodology log)
Phase 2.5: Batch Source Hashing
After all Phase 2 agents complete, deploy a general-purpose batch hashing agent. Agent FIRST ACTION: Read ${CLAUDE_SKILL_DIR}/provenance.md and execute the batch hashing protocol.
Phase 3: Research Synthesis
Deploy synthesis-specialist:
- FIRST ACTION: Read
BASE_DIR/[TOPIC_SLUG]_Research_Outline.md
- Read ALL agent output: claims tables (
_Claims_[AgentID].md), bibliographies (_[AgentID]_Bibliography.md), sources (_Sources_[AgentID].md), methodology logs (_Methodology_Log_[AgentID].md)
- Consolidate per-agent methodology logs into
BASE_DIR/[TOPIC_SLUG]_Methodology_Log.md
- Consolidate per-agent sources into
BASE_DIR/[TOPIC_SLUG]_Sources.md
- Integrate findings, cross-reference claims, resolve contradictions, mark confidence scores
- Structure analysis for strategic IP decisions: landscape overview, filing trends, competitive matrices, FTO risk ratings, whitespace opportunities
- Save to
BASE_DIR/[TOPIC_SLUG]_Synthesis_Report.md
Phase 4: Conditional Reporting
If --no-vvc flag is present: Phase 4 heading is "Professional Reporting". Output file: _Comprehensive_Report.md. Skip claim tagging. Phases 5-6 are skipped.
Otherwise (VVC enabled, no --no-vvc flag): Phase 4 heading is "Draft Reporting". Output file: _Draft_Report.md. Apply claim tagging per ${CLAUDE_SKILL_DIR}/vvc-pipeline.md.
Deploy research-reporting-specialist: transform synthesis into professional report. Tone: Professional patent intelligence assessment for IP attorneys, technology transfer officers, and C-suite. Sections: Executive Summary, Patent Landscape Overview, Key Patent Families, Claims Analysis, Prior Art Assessment, Competitive Portfolio Matrix, Freedom-to-Operate Assessment, Whitespace & Opportunity Analysis, IP Risk Matrix, Recommendations, Methodology, Bibliography. Use APA 7th Edition with patent extensions. Consolidate master bibliography.
Phase 4.5: Provenance Audit
Tier behavior: Quick: skip | Standard/Deep/Comprehensive: run. Deploy a general-purpose provenance audit agent. Agent FIRST ACTION: Read ${CLAUDE_SKILL_DIR}/provenance.md and execute the provenance audit protocol.
Phase 5-6: VVC Verification & Correction
VVC specialist reads ${CLAUDE_SKILL_DIR}/vvc-pipeline.md for full instructions. If NO_VVC is true, skip entirely. Now executing research deployment...
File Output Structure
./research-reports/${RUN_TS}_${TOPIC_SLUG}/
├── _status/
│ ├── _pipeline.json
│ ├── [AgentID].json # Per-agent status
│ ├── [AgentID]_claims.json # Per-agent claims (new)
│ ├── [AgentID]_log.json # Per-agent action log (new)
│ ├── vvc-specialist_verification.json # VVC verdicts (new, Phase 5-6)
│ ├── server.js # Dashboard server
│ └── dashboard.html # Dashboard UI
├── [TOPIC_SLUG]_Research_Outline.md
├── [TOPIC_SLUG]_Claims_[AgentID].md # Per-agent
├── [TOPIC_SLUG]_Sources_[AgentID].md # Per-agent
├── [TOPIC_SLUG]_Methodology_Log_[AgentID].md # Per-agent
├── [TOPIC_SLUG]_[AgentID]_Bibliography.md # Per-agent
├── [TOPIC_SLUG]_Methodology_Log.md # Consolidated
├── [TOPIC_SLUG]_Sources.md # Consolidated
├── [TOPIC_SLUG]_Synthesis_Report.md
├── [TOPIC_SLUG]_Draft_Report.md
├── [TOPIC_SLUG]_VVC_Verification_Report.md # Phase 5
├── [TOPIC_SLUG]_VVC_Correction_Log.md # Phase 6 (Deep/Comprehensive)
├── [TOPIC_SLUG]_Hash_Manifest.md # Provenance
├── [TOPIC_SLUG]_Provenance_Log.md # Provenance audit
├── [TOPIC_SLUG]_Comprehensive_Report.md # Final output
└── [TOPIC_SLUG]_Master_Bibliography.md
Key Workflow Features
- Planning-first coordination -- all agents read the outline as their first action
- Iterative Search-Assess-Refine -- up to 4 refinement passes per question
- Per-agent file isolation -- each agent writes to its own files; synthesis consolidates
- Confidence scoring on every claim (HIGH/MEDIUM/LOW/SPECULATIVE)
- Source credibility hierarchy -- no HIGH confidence claim on Tier 4-5 sources alone
- Domain specialization -- all agents operate within IP and patent landscape analysis context
- Cryptographic provenance -- batch hashing after Phase 2;
--reverifiable retains snapshots
- VVC pipeline -- two-pass verify+correct with
--no-vvc opt-out
- Claim type taxonomy -- [VC]/[PO]/[IE] tagging focuses verification effort
Domain Preamble
You are conducting patent intelligence analysis to the standard expected by a senior IP attorney or head of technology transfer. Accuracy of patent data is paramount: every patent number, filing date, assignee, and claim reference must be verified against official records. Distinguish granted patents (enforceable rights) from published applications (potential future rights). Assess patent strength via claim breadth, prosecution history estoppel, geographic coverage, and remaining term.
All agents apply IP and patent landscape analysis-specific knowledge, terminology, and analytical frameworks for IP attorneys, technology transfer officers, and R&D strategists.
Operational Lessons
No entries yet -- update after first research run with /post-mortem.