### Install Seine Plugin (Manual) Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/docs/llms.txt Instructions for manually cloning the Seine agentic search orchestrator plugin repository from GitHub. ```bash git clone https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin ``` -------------------------------- ### Install and Execute Seine Search Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/README.md Commands to install the Seine plugin via the marketplace and execute a search query with a specific depth level. ```bash # Install /plugin marketplace add adambkovacs/seine-agentic-search-orchestrator-plugin /plugin install seine # Run /seine:seine-search "Impact of EU AI Act on startups" dig ``` -------------------------------- ### Install Seine Plugin with Claude Code Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/landing/index.html Instructions for installing the Seine plugin using the Claude Code command-line interface. This involves adding the plugin to the marketplace and then installing it. ```bash /plugin marketplace add adambkovacs/seine-agentic-search-orchestrator-plugin /plugin install seine ``` -------------------------------- ### Install Seine Plugin (Claude Code) Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/docs/llms.txt Command to install the Seine plugin directly within the Claude Code environment. ```bash /plugin install seine ``` -------------------------------- ### Artifact Creation and Writing (JavaScript) Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt Provides JavaScript code examples for interacting with the file system to create artifact directories and write initial query data. It uses hypothetical `Bash` and `Write` functions to simulate command execution and file writing. ```javascript // Create artifact directory Bash({ command: "mkdir -p research/artifacts/eu-ai-act-analysis-2026-03-02/{01-search-rounds,03-triage,04-council-r1,05-research}" }) // Write initial query artifact Write({ file_path: "research/artifacts/eu-ai-act-analysis-2026-03-02/00-query.json", content: JSON.stringify({ query: "EU AI Act impact on fintech startups", depth: "drill", domains_selected: ["web", "academic", "osint"], routing_decision: "Query involves regulatory compliance requiring official sources (osint), academic analysis, and general coverage", timestamp: "2026-03-02T14:30:00Z" }) }) ``` -------------------------------- ### WebSearch Query Construction Patterns (JavaScript) Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt Demonstrates how to construct effective multi-domain search queries using the Claude Code's native WebSearch tool. It includes examples for the 3-Query Framework, domain-specific patterns (academic, OSINT, social media, regulatory), and various boolean operators for refining searches. ```javascript const queries = { // Query 1: Direct query direct: '"EU AI Act" "fintech compliance" requirements 2026', // Query 2: Angle shift (different framing) angleShift: 'site:linkedin.com/jobs "AI compliance" OR "regulatory" fintech Berlin', // Query 3: Counter-evidence (what would disprove the claim) counterEvidence: '"EU AI Act" "delayed" OR "postponed" OR "exemption" fintech' }; // Domain-specific search patterns const domainPatterns = { // Academic sources academic: 'site:arxiv.org "transformer architecture" 2026', // OSINT - SEC filings edgar: 'site:sec.gov "Company Name" "10-K" OR "10-Q" OR "DEF 14A"', // OSINT - Corporate registry openCorporates: 'site:opencorporates.com "Company Name"', // OSINT - Sanctions ofac: 'site:treasury.gov "Entity Name" "sanctions"', // LinkedIn X-ray (bypass login) linkedin: 'site:linkedin.com/in "CEO" OR "CTO" "[Company]"', // Social media sentiment reddit: 'site:reddit.com "[Product]" "switched" OR "alternative" OR "honest review"', // Government/regulatory euLaw: 'site:eur-lex.europa.eu "[Topic]" "regulation" OR "directive"' }; // Boolean operators const operators = { exactPhrase: '"EU AI Act Article 4"', siteRestrict: 'site:arxiv.org "topic"', booleanOr: '"AI training" OR "AI enablement"', exclude: '"AI training" -chatgpt -tutorial', grouping: '(grant OR funding) "AI" site:europa.eu', fileType: 'filetype:pdf "strategic plan" 2026', dateRange: '"AI regulation" after:2025-06-01', proximity: '"Delivery Hero" AROUND(3) "AI training"' }; ``` -------------------------------- ### Research Report Output Format Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt An example of the final rendered markdown output produced by the Seine pipeline, featuring structured headers, inline citations, evidence labels, and confidence summary tables. ```markdown # EU AI Act vs US AI Regulation: Comparative Analysis > Pipeline: drill | 4 search rounds | 34 sources | 7 council members > Evidence: SOLID (verified) | SOFT (single source) | SHAKY (weak) | UNKNOWN > Artifacts: research/artifacts/eu-ai-act-analysis-2026-03-02/ The EU AI Act classifies AI systems into four risk tiers [1] [Source-Critic: SOLID]: unacceptable (banned outright), high-risk (strict compliance), limited (transparency obligations), and minimal (no specific rules). The EU AI Act's GPAI 10^25 FLOPs threshold operates as a rebuttable presumption, not a hard ceiling, but the adversarial reviewer found that the inverse burden of proof makes practical rebuttal difficult. [4] [Adversarial: SOFT] The Trump executive order's claim to preempt state AI laws was downgraded from SOFT to SHAKY by the adversarial reviewer. The EO lacks self-executing legal authority. [6] [9] ## Sources & References | # | Title | Trust | Cited By | |---|-------|-------|----------| | 1 | EUR-Lex: Regulation 2024/1689 | HIGH | hunter, skeptic, referee | | 2 | EU AI Office: Implementation Timeline | HIGH | hunter, scout | ## Confidence Summary | Claim | Label | Sources | Score | |-------|-------|---------|-------| | 4-tier risk classification | SOLID | 8 | 0.92 | | No US federal comprehensive AI law | SOLID | 5 | 0.87 | | Trump EO federal preemption | SHAKY | 3 | 0.34 | ``` -------------------------------- ### Dispatching Parallel Domain Searches with WebSearch Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-search/SKILL.md This snippet demonstrates how to initiate parallel searches across multiple domains using the `WebSearch` tool within the Seine orchestrator. It's designed for the 'Dispatch' step, where selected domains are queried simultaneously. The example shows how to construct search queries tailored for specific domains like academic and social media. ```Claude Code Agent(seine-research-hunter) ← academic search WebSearch(query + "site:arxiv.org") WebSearch(query + site:reddit.com OR site:twitter.com) ``` -------------------------------- ### Initialize Artifact Directory Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-search/SKILL.md Creates the directory structure for storing research artifacts based on a generated slug. This ensures a consistent file system hierarchy for search rounds and metadata. ```javascript Bash({ command: "mkdir -p research/artifacts/{slug}/01-search-rounds" }) ``` -------------------------------- ### Initialize Artifact Directory Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/agents/seine-researcher.md Creates the required directory structure for research artifacts using the Bash tool. This must be the first step before spawning any agents or invoking skills. ```javascript Bash({ command: "mkdir -p research/artifacts/{slug}/{01-search-rounds,03-triage,04-council-r1,05-research}" }) ``` -------------------------------- ### Create Council Directory (Bash) Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-council/SKILL.md Creates a new directory for council R1 or R2 artifacts. This command ensures the directory exists, creating it if it doesn't. It utilizes the Bash command `mkdir -p` for this purpose. ```javascript Bash({ command: "mkdir -p {artifact_dir}/04-council-r1" }) Bash({ command: "mkdir -p {artifact_dir}/06-council-r2" }) ``` -------------------------------- ### Configure Chart.js Instances for Seine Theme Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/agents/seine-report-renderer.md Initializes a Chart.js instance with the required Seine dark theme, including specific grid colors, font settings, and custom tooltip styling. ```javascript new Chart(ctx, { type: 'bar', data: { labels: [...], datasets: [{ data: [...], backgroundColor: [...], borderRadius: 4 }] }, options: { indexAxis: 'y', scales: { x: { max: 1.0, grid: { color: 'rgba(255,255,255,0.06)' }, ticks: { color: '#718096' } } }, plugins: { legend: { display: false }, tooltip: { backgroundColor: '#12121A', titleColor: '#E2E8F0' } } } }); ``` -------------------------------- ### Create Research Directory - JavaScript Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-research/SKILL.md Creates the research directory structure using `mkdir -p` if the `artifact_dir` is provided. This command ensures that the directory and any necessary parent directories are created without errors if they already exist. ```javascript Bash({ command: "mkdir -p {artifact_dir}/05-research" }) ``` -------------------------------- ### Define Output Header Format Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/agents/seine-kb/REFERENCE.md The mandatory header format for rendered research documents, providing metadata about the pipeline depth, search rounds, source counts, and artifact locations. ```markdown > Pipeline: [depth] | [N] search rounds | [N] sources | [N] council members > Evidence vocabulary: SOLID | SOFT | SHAKY | UNKNOWN > Artifacts: research/artifacts/{slug}/ ``` -------------------------------- ### Spawn Output Renderer Agent Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-search/SKILL.md Invokes the seine-output-renderer agent to compile JSON artifacts into a structured markdown document containing citations, source tables, and work logs. ```javascript Agent({ subagent_type: "seine-output-renderer", model: "sonnet", prompt: "Artifact directory: research/artifacts/{slug}/\nQuery: \nDepth: \n\nInstructions:\n1. Read all JSON files in the artifact directory (00-query.json, 01-search-rounds/*.json, 02-fusion.json, and any 03-triage/, 04-council-r1/, 05-research/, 06-council-r2/, 07-sources.json files that exist).\n2. Produce a markdown document containing:\n - A prose body with inline [N] citations referencing numbered sources\n - A Sources & References table: | # | Title | URL | Trust Tier | Cited By |\n - A Work Log section listing every search round, triage verdict, council member, and research phase with timestamps\n - A Confidence Summary section with per-claim confidence scores from the evidence vocabulary\n3. Return the complete markdown document." }) ``` -------------------------------- ### Run Gated Research Pipeline Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt Executes a three-phase research pipeline involving discovery, analysis, and synthesis agents. The process includes validation checkpoints that halt execution if quality thresholds are not met. ```bash # Standard deep research with Sonnet models /seine:seine-research "State of open-source LLMs for enterprise deployment" drill ``` -------------------------------- ### Render Interactive Research Dossiers Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt Converts research artifacts into self-contained HTML dossiers with integrated Chart.js visualizations. Supports both CLI-based rendering and programmatic agent-based execution. ```bash # Render from specific artifact directory /seine:seine-render research/artifacts/eu-ai-act-analysis-2026-03-02 # Render most recent research (auto-detects latest) /seine:seine-render ``` ```javascript // Spawn report renderer agent Agent({ subagent_type: "seine-report-renderer", model: "sonnet", prompt: "Artifact directory: research/artifacts/eu-ai-act-analysis-2026-03-02\n\nRead all JSON files in the artifact directory (00-query.json through 08-timeline.json, skip missing). Auto-detect chartable data from the findings. Generate a self-contained HTML dossier with Chart.js charts bundled inline. Write the output to research/artifacts/eu-ai-act-analysis-2026-03-02/09-dossier.html." }) // Verify output Bash({ command: "ls -la research/artifacts/eu-ai-act-analysis-2026-03-02/09-dossier.html" }) ``` -------------------------------- ### Run Humanizer Agent Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/agents/seine-researcher.md Initializes and runs the 'seine-humanizer' agent. It takes the model type and a prompt that specifies the document path, type, and quality gate criteria. The output is saved to a new markdown file. ```javascript Agent({ subagent_type: "seine-humanizer", model: "sonnet", prompt: "Document path: research/artifacts/{slug}/08-rendered.md\nDocument type: \nApply anti-slop detection (5 tiers), voice application, and quality gate (NO-SLOP > 90%)." }) ``` -------------------------------- ### Implement Clickable Citations Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/agents/seine-report-renderer.md Provides the HTML structure for inline citations and the corresponding CSS target highlighting for the source list. ```html [1]
[1] Title HIGH
``` ```css .source:target { background: rgba(99, 179, 237, 0.08); border-left: 3px solid var(--accent); transition: background 0.4s; } ``` -------------------------------- ### Artifact Directory Structure (Bash) Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt Defines a persistent output structure for maintaining a full audit trail of research artifacts. This structure organizes files by stage, from initial queries and search rounds to fusion, triage, council reviews, and final reports. ```bash research/artifacts/{query-slug}-{date}/ ├── 00-query.json # Original query + routing decision + domains ├── 01-search-rounds/ # Per-domain raw results │ ├── round-1-web.json │ ├── round-1-academic.json │ └── round-2-osint.json ├── 02-fusion.json # RRF-fused results with scores ├── 03-triage/ # 3 triage verdicts │ ├── completeness.json │ ├── quality.json │ └── gaps.json ├── 04-council-r1/ # 7 council member outputs │ ├── synthesizer.json │ ├── contrarian.json │ ├── lateral-hunter.json │ ├── source-critic.json │ ├── pattern-spotter.json │ ├── blind-spot.json │ └── temporal.json ├── 05-research/ # Research phase outputs │ ├── phase-a-hunter.json │ ├── phase-a-scout.json │ ├── gate-a.json │ ├── phase-b-skeptic.json │ ├── phase-b-referee.json │ ├── gate-b.json │ ├── phase-c-adversarial.json │ └── phase-c-confidence.json ├── 06-council-r2/ # R2 member outputs (if run) ├── 07-sources.json # Deduplicated master source list ├── 08-timeline.json # Per-stage timing └── 09-dossier.html # Interactive HTML report ``` -------------------------------- ### Execute Multi-Domain Search with Seine Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt Orchestrates searches across web, academic, and OSINT domains using depth-controlled levels. Users can trigger different intensities ranging from quick 'skim' operations to exhaustive 'siege' research. ```bash # Quick fact-check (route + fuse only, ~30 seconds) /seine:seine-search "Latest transformer architecture papers" skim # Surface-level overview with triage gate (~1-2 minutes) /seine:seine-search "EU AI Act compliance requirements for SMEs" scan # Thorough analysis with 7-member council deliberation (~3-5 minutes) /seine:seine-search "Impact of EU AI Act on startups" dig # Deep investigation with full research pipeline (~8-15 minutes) /seine:seine-search "State of open-source LLMs for enterprise deployment" drill # Exhaustive multi-round research with Opus models (~20-40 minutes) /seine:seine-search "China's semiconductor strategy and US export controls" siege ``` -------------------------------- ### Initialize Research Subagents Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-research/SKILL.md Spawns specialized research agents (hunter, scout, skeptic, referee, etc.) using the Agent tool. These calls require dynamic model selection based on the research depth (drill or siege) and specific prompt instructions. ```javascript Agent({ subagent_type: "seine-research-hunter", model: "", prompt: "Query: \nPrior results: \nTask: Build evidence map and confidence table." }); Agent({ subagent_type: "seine-research-scout", model: "", prompt: "Query: \nPrior results: \nTask: Find adjacent signals, weak signals, and timing triggers." }); ``` -------------------------------- ### Update Plugin via Marketplace Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/README.md Command to update the Seine plugin using the marketplace. This is the recommended method for keeping the plugin up-to-date. ```bash # Update via marketplace /plugin update seine ``` -------------------------------- ### Spawn Deliberate Agents - Round 1 (JavaScript) Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-council/SKILL.md Spawns seven council members (synthesizer, contrarian, lateral-hunter, source-critic, pattern-spotter, blind-spot, temporal) in parallel for the first round of multi-perspective analysis. Each member processes a query and search results, returning findings and a verdict. ```javascript Agent({ subagent_type: "seine-council-synthesizer", model: "sonnet", prompt: "Query: \nSearch results: " }) Agent({ subagent_type: "seine-council-contrarian", model: "sonnet", prompt: "Query: \nSearch results: " }) Agent({ subagent_type: "seine-council-lateral-hunter", model: "sonnet", prompt: "Query: \nSearch results: " }) Agent({ subagent_type: "seine-council-source-critic", model: "sonnet", prompt: "Query: \nSearch results: " }) Agent({ subagent_type: "seine-council-pattern-spotter", model: "sonnet", prompt: "Query: \nSearch results: " }) Agent({ subagent_type: "seine-council-blind-spot", model: "sonnet", prompt: "Query: \nSearch results: " }) Agent({ subagent_type: "seine-council-temporal", model: "sonnet", prompt: "Query: \nSearch results: " }) ``` -------------------------------- ### Verify dossier output Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-render/SKILL.md Executes a Bash command to confirm the existence and file size of the generated dossier. This ensures the rendering process completed successfully before reporting back to the user. ```javascript Bash({ command: "ls -la {artifact_dir}/09-dossier.html && wc -c {artifact_dir}/09-dossier.html" }) ``` -------------------------------- ### Evidence Vocabulary and Confidence Calculation (JavaScript) Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt Provides the universal 4-level confidence labeling system (SOLID, SOFT, SHAKY, UNKNOWN) and associated numeric values. Includes functions for calculating composite confidence scores based on evidence, source quality, recency, and agreement, and converting scores to labels. ```javascript // Evidence labels and their numeric values const EVIDENCE_LABELS = { SOLID: 1.0, // Multiple independent sources, no contradictions SOFT: 0.6, // Single credible source or indirect evidence SHAKY: 0.3, // Single biased source or conflicting evidence UNKNOWN: 0.0 // Insufficient evidence to assess }; // Confidence formula const calculateConfidence = (evidence, sourceQuality, recency, agreement) => { return (evidence * 0.40) + (sourceQuality * 0.25) + (recency * 0.20) + (agreement * 0.15); }; // Composite score to label conversion const getEvidenceLabel = (score) => { if (score >= 0.80) return "SOLID"; if (score >= 0.55) return "SOFT"; if (score >= 0.30) return "SHAKY"; return "UNKNOWN"; }; // Source quality tiers const SOURCE_QUALITY = { HIGH: "Peer-reviewed, .gov, official records", // Can support SOLID MEDIUM: "Major news (Reuters, Bloomberg), reports", // Supports SOFT with corroboration LOW: "Forums, unvetted blogs, social media", // SHAKY at best DISQUALIFIED: "Anonymous, content farms, retracted" // Always UNKNOWN }; // Recency tiers const RECENCY = { CURRENT: "< 30 days", // Highest weight RECENT: "30-180 days", DATED: "180-365 days", STALE: "> 365 days" // Lowest weight }; ``` -------------------------------- ### Writing the HTML Output File Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/agents/seine-report-renderer.md This JavaScript snippet demonstrates how the final, complete HTML content is written to a specified file path. It takes the generated HTML string and saves it as '09-dossier.html' within the artifact directory. ```javascript Write({ file_path: "{artifact_dir}/09-dossier.html", content: "" }) ``` -------------------------------- ### Execute Seine Output Rendering Pipeline Source: https://context7.com/adambkovacs/seine-agentic-search-orchestrator-plugin/llms.txt A three-stage agentic workflow that transforms raw JSON research artifacts into polished markdown prose, performs an anti-slop audit for voice quality, and generates an interactive HTML dossier with embedded charts. ```javascript // Stage 1: Output Renderer - transforms JSON to prose Agent({ subagent_type: "seine-output-renderer", model: "sonnet", prompt: `Artifact directory: research/artifacts/eu-ai-act-analysis-2026-03-02/ Query: EU AI Act impact on fintech Depth: drill Instructions: 1. Read all JSON files in the artifact directory 2. Produce a markdown document containing: - A prose body with inline [N] citations referencing numbered sources - [Member: LABEL] attributions for council-challenged findings - A Sources & References table - A Work Log section - A Confidence Summary section 3. Return the complete markdown document.` }) // Stage 2: Humanizer - anti-slop audit and voice styling Agent({ subagent_type: "seine-humanizer", model: "sonnet", prompt: `Document to humanize: Document type: research Instructions: 1. Run 5-tier anti-slop audit 2. Apply voice settings appropriate to document type 3. Compute NO-SLOP score (must exceed 90%) 4. Return polished markdown and final score.` }) // Stage 3: Report Renderer - interactive HTML dossier Agent({ subagent_type: "seine-report-renderer", model: "sonnet", prompt: `Artifact directory: research/artifacts/eu-ai-act-analysis-2026-03-02/ Read all JSON files. Auto-detect chartable data. Generate self-contained HTML dossier with Chart.js charts bundled inline. Write to research/artifacts/eu-ai-act-analysis-2026-03-02/09-dossier.html.` }) ``` -------------------------------- ### Compile and Write Sources - JavaScript Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-research/SKILL.md Compiles a master list of unique sources from all agent outputs after Phase C. It deduplicates sources by URL, computes fields like trust tier and citation count, sorts them, and writes the result to `07-sources.json`. The `Write` tool is used for persistence. ```javascript Write({ file_path: "{artifact_dir}/07-sources.json", content: "" }) ``` -------------------------------- ### Generate HTML Dossier Source: https://github.com/adambkovacs/seine-agentic-search-orchestrator-plugin/blob/main/skills/seine-search/SKILL.md Invokes the seine-report-renderer to create a self-contained HTML report with embedded Chart.js visualizations based on the collected research artifacts. ```javascript Agent({ subagent_type: "seine-report-renderer", model: "sonnet", prompt: "Artifact directory: research/artifacts/{slug}/\n\nRead all JSON files in the artifact directory (00-query.json through 08-timeline.json, skip missing). Auto-detect chartable data from the findings. Generate a self-contained HTML dossier with Chart.js charts bundled inline. Write the output to research/artifacts/{slug}/09-dossier.html." }) ```