id: analyze-paper-v1 name: analyze-paper version: 1.0.0 author: raahul@getbindu.com

Description

description: | Analyzes truth claims and arguments using LLM-based reasoning. Extracts claims from input text, provides supporting and refuting evidence, identifies logical fallacies, assigns quality ratings (A-F), and generates balanced assessments. Uses OpenRouter LLM with ArxivTools for academic paper access.

Tags and Modes

tags:

• truth-claims
• argument-analysis
• critical-thinking
• fact-checking
• logical-fallacies

input_modes:

• text/plain
• application/json

output_modes:

• text/plain
• application/json

Example Queries

examples:

• "Analyze the truth claims in this argument about climate change"
• "Evaluate the claims and provide supporting and refuting evidence"
• "Identify logical fallacies and rate the quality of these claims"

Detailed Capabilities

capabilities_detail: primary_capability: supported: true description: "LLM-based truth claim analysis with centrist orientation" features: - "Extracts truth claims from input text (< 16 words per claim)" - "Provides claim support evidence via LLM reasoning" - "Provides claim refutation evidence via LLM reasoning" - "Identifies logical fallacies with quoted examples" - "Assigns quality ratings (A/B/C/D/F scale)" - "Generates argument summary (< 30 words)" - "Calculates lowest, highest, and average claim scores" - "Provides characterization labels (e.g., specious, extreme-right, weak, baseless)" - "Generates overall analysis with recommendations (30 words)" limitations: "Evidence quality depends on LLM knowledge cutoff; ArxivTools access limited to arXiv papers; no real-time fact verification API"

secondary_capability: supported: true description: "ArxivTools integration for academic paper retrieval" features: - "Can search and retrieve papers from arXiv" - "Access to arXiv metadata and abstracts" limitations: "Only arXiv papers; no direct access to other academic databases"

Requirements

requirements: packages: - "agno" - "bindu" - "python-dotenv" apis: - "OPENROUTER_API_KEY (required)" - "MEM0_API_KEY (required)" system: - "Internet access for ArxivTools" min_memory_mb: 256

Performance Metrics

performance: avg_processing_time_ms: 30000 max_concurrent_requests: 10 memory_per_request_mb: 256 scalability: horizontal

Tool Restrictions

allowed_tools:

• ArxivTools

Rich Documentation

documentation: overview: | This skill provides centrist-oriented analysis of truth claims and arguments using LLM reasoning. It extracts claims, provides both supporting and refuting evidence, identifies logical fallacies, and assigns quality ratings. Built on Agno framework with OpenRouter LLM (default: openai/gpt-5.2-chat). Includes ArxivTools for accessing academic papers from arXiv. Uses Mem0 for memory (API key required). Exposes JSON-RPC 2.0 API via bindufy for task submission and retrieval.

use_cases: when_to_use: - "Analyzing truth claims in arguments or text" - "Getting balanced perspectives with supporting and refuting evidence" - "Identifying logical fallacies in arguments" - "Rating claim quality on A-F scale" - "Centrist-oriented critical analysis"

when_not_to_use:
  - "Real-time fact verification (LLM has knowledge cutoff)"
  - "Summarization only (use summarization skill)"
  - "Creative writing or content generation"
  - "Specialized domain analysis requiring recent data beyond LLM training"

input_structure: | Accepts text containing arguments or claims via JSON-RPC 2.0 message/send method.

Request format:
{
  "jsonrpc": "2.0",
  "method": "message/send",
  "params": {
    "message": {
      "role": "user",
      "parts": [{"kind": "text", "text": "[argument or claims to analyze]"}],
      "kind": "message",
      "messageId": "[uuid]",
      "contextId": "[uuid]",
      "taskId": "[uuid]"
    },
    "skillId": "analyze-paper-v1",
    "configuration": {"acceptedOutputModes": ["application/json"]}
  },
  "id": "[request-id]"
}

Constraints:
- Input should contain arguments or claims
- No hard length limits enforced in code
- English text recommended

output_format: | Returns JSON-RPC 2.0 response with task submission, then markdown-formatted analysis via tasks/get.

Initial response (message/send):
{
  "jsonrpc": "2.0",
  "id": "[request-id]",
  "result": {
    "id": "[task-id]",
    "status": {"state": "submitted", "timestamp": "[iso-timestamp]"},
    "history": [...]
  }
}

Task result (tasks/get):
{
  "result": {
    "status": {"state": "completed"},
    "history": [{"role": "assistant", "parts": [{"kind": "text", "text": "[markdown analysis]"}]}],
    "artifacts": [{"parts": [{"kind": "text", "text": "[markdown analysis]"}]}]
  }
}

Analysis structure (markdown):
- ARGUMENT SUMMARY: (< 30 words)
- TRUTH CLAIMS: For each claim:
  - CLAIM: (< 16 words)
  - CLAIM SUPPORT EVIDENCE:
  - CLAIM REFUTATION EVIDENCE:
  - LOGICAL FALLACIES:
  - CLAIM RATING: A/B/C/D/F
  - LABELS:
- OVERALL SCORE:
  - LOWEST CLAIM SCORE:
  - HIGHEST CLAIM SCORE:
  - AVERAGE CLAIM SCORE:
- OVERALL ANALYSIS: (30 words)

error_handling: - "Missing API keys: Raises ValueError on startup" - "Agent not initialized: Raises RuntimeError if handler called before init" - "LLM errors: Propagated from OpenRouter API" - "ArxivTools errors: Propagated from tool execution" - "No explicit input validation in code"

examples: - title: Transformer Paper Analysis input: text: Analyze the Transformer architecture paper and provide structured evaluation with methodology assessment and quality scoring output: format: Markdown with sections for Executive Summary, Methodology Evaluation, Results and Statistical Validity, Strengths, Weaknesses, Reproducibility, and Overall Quality Score processing_time_ms: 30000 note: Output format varies based on LLM interpretation and may not strictly follow TRUTH CLAIMS template

- title: Clinical Trial Analysis
  input:
    text: Critically analyze a randomized controlled trial for diabetes medication with assessment of study design and statistical methods
  output:
    format: Markdown with sections for Study Design, Randomization Quality, Statistical Methods, Risk of Bias, External Validity, and Scientific Strength Rating
    processing_time_ms: 28000
    note: LLM adapts output structure to request and may not follow exact TRUTH CLAIMS format

best_practices: for_developers: - "Use JSON-RPC 2.0 message/send to submit tasks" - "Poll tasks/get to retrieve completed results" - "Provide clear arguments or claims in input text" - "Allow 30+ seconds for LLM processing" - "Set acceptedOutputModes to application/json in configuration"

for_orchestrators:
  - "Route truth claim analysis and argument evaluation here"
  - "Task state transitions: submitted → completed"
  - "Results available in both history and artifacts"
  - "No built-in retry logic; implement externally"
  - "Monitor for LLM API availability and rate limits"

installation: | Environment variables required: - OPENROUTER_API_KEY: Get from https://openrouter.ai/keys (required) - MEM0_API_KEY: Get from https://app.mem0.ai/dashboard/api-keys (required) - MODEL_NAME: Optional, defaults to openai/gpt-5.2-chat

Quick start:
1. Set environment variables in .env file
2. Install dependencies: uv sync
3. Start agent: python -m analyze_paper_agent.main
4. Agent listens on port defined in agent_config.json
5. Send JSON-RPC 2.0 requests to http://localhost:[port]

Command-line options:
--model: Override MODEL_NAME
--api-key: Override OPENROUTER_API_KEY
--mem0-api-key: Override MEM0_API_KEY

versioning: - version: "1.0.0" date: "2026-03-02" changes: "Initial release with LLM-based truth claim analysis, ArxivTools integration, JSON-RPC 2.0 API via bindufy"

Assessment fields for skill negotiation

assessment: keywords: - "truth claims" - "argument analysis" - "logical fallacy" - "evidence" - "claim rating" - "centrist analysis" - "supporting evidence" - "refuting evidence" - "claim verification"

specializations: - domain: "truth claim analysis" confidence_boost: 0.5 - domain: "argument evaluation" confidence_boost: 0.4 - domain: "logical fallacy detection" confidence_boost: 0.4 - domain: "balanced assessment" confidence_boost: 0.3

anti_patterns: - "summarize without analysis" - "write a paper" - "translate text" - "generate creative content" - "real-time fact verification"

complexity_indicators: simple: - "single claim with evidence" - "basic argument evaluation" medium: - "multiple claims with fallacy detection" - "argument with supporting and refuting evidence" complex: - "comprehensive analysis with 10+ claims" - "multi-faceted arguments with detailed ratings"