메뉴
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.
| 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. |
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.
/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)| 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 |
Parse tier from $ARGUMENTS: --quick / --standard / --deep / --comprehensive (default: Standard). Additional flags: --outline-only (stop after Phase 1), --approve / --no-approve (user gate control), --extend (project-aware context), --reverifiable (retain source snapshots), --no-vvc (skip Phases 5-6, Phase 4 becomes final reporting).
--no-vvc is present, set NO_VVC to true (skip Phases 5-6, Phase 4 becomes final reporting)http:// or https://, treat as URL seed — fetch, extract topic, derive slug from titleTOPIC_SLUG = lowercase topic, hyphenate spaces, strip punctuation
RUN_TS = YYYY-MM-DD_HHMMSS_ET (Eastern Time via bash: TZ='America/New_York' date '+%Y-%m-%d_%H%M%S_ET')
BASE_DIR = "./research-reports/${RUN_TS}_${TOPIC_SLUG}"
ENGINE_ID = "patent-intelligence-engine"
Create BASE_DIR/_status/ directory. Write BASE_DIR/_status/_pipeline.json with initial pipeline state. Substitute all runtime variables (ENGINE_ID, TOPIC_SLUG, detected tier, agent IDs for this tier):
{
"engineId": "${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.
Before each phase begins, update _pipeline.json:
currentPhase and phaseLabel to the new phase"in_progress"startedAt to the current ISO-8601 timestamplastUpdated timestampAfter each phase completes:
"complete"completedAt to the current ISO-8601 timestamplastUpdated timestampSkip this step for Quick tier.
${CLAUDE_SKILL_DIR}/dashboard-server.js and ${CLAUDE_SKILL_DIR}/dashboard.html to BASE_DIR/_status/node BASE_DIR/_status/server.js in background via Bash tool_pipeline.json with dashboardUrlAll files live under BASE_DIR. Never restate the prompt in chat; write long data to files and keep chat responses concise.
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.
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.
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.
Deploy research agents per tier config. Each agent operates independently. Each agent writes to its own files only.
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.
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.
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:
${CLAUDE_SKILL_DIR}/standards.md, ${CLAUDE_SKILL_DIR}/research-protocol.md, and BASE_DIR/[TOPIC_SLUG]_Research_Outline.md_[AgentID]_Bibliography.md), claims table (_Claims_[AgentID].md), methodology log (_Methodology_Log_[AgentID].md), sources (_Sources_[AgentID].md)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:
Decision logic:
Log verification results to BASE_DIR/[TOPIC_SLUG]_Methodology_Log.md (this becomes the first entry in the consolidated methodology log)
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.
Deploy synthesis-specialist:
BASE_DIR/[TOPIC_SLUG]_Research_Outline.md_Claims_[AgentID].md), bibliographies (_[AgentID]_Bibliography.md), sources (_Sources_[AgentID].md), methodology logs (_Methodology_Log_[AgentID].md)BASE_DIR/[TOPIC_SLUG]_Methodology_Log.mdBASE_DIR/[TOPIC_SLUG]_Sources.mdBASE_DIR/[TOPIC_SLUG]_Synthesis_Report.mdIf --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.
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.
VVC specialist reads ${CLAUDE_SKILL_DIR}/vvc-pipeline.md for full instructions. If NO_VVC is true, skip entirely. Now executing research deployment...
./research-reports/${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
--reverifiable retains snapshots--no-vvc opt-outYou 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.
No entries yet -- update after first research run with /post-mortem.