- Home
- Skills
- Agents & Orchestration
- mine2digest
mine2digest
A systematic literature review pipeline for dual-agent screening, evidence extraction, and auditable draft verification.
Free
Works with the AI Agents you already use
mine2digest
Example session with this skill installed
Initialize a new literature project for "CLIP multimodal generalization", add 10 candidates from OpenAlex, and show me the screening queue.
- Read your context and instructions
- Compiled the mine2digest
I have initialized the workspace and populated the candidate database with 10 records from OpenAlex regarding CLIP multimodal generalization.
Workspace Initialization & Discovery
| Step | Action | Status | Result |
|---|---|---|---|
| 1 | mine2digest init | Success | Workspace m2d/ created |
| 2 | openalex_seed.py | Success | 10 candidates retrieved |
| 3 | add-candidates | Success | 10 records inserted to SQLite |
| 4 | next-candidates | Pending | 10 papers awaiting dual-agent votes |
Current Screening Queue (Top 5)
| ID | Title | Source | Year | Status |
|---|---|---|---|---|
oa:W4289... | Learning Transferable Visual Models... | openalex | 2021 | candidate |
oa:W3162... | Robust Fine-tuning of Zero-shot Models | openalex | 2022 | candidate |
oa:W4301... | Multimodal Few-Shot Learning with CLIP | openalex | 2023 | candidate |
oa:W2987... | Evaluating CLIP on Out-of-Distribution... | openalex | 2021 | candidate |
oa:W3210... | Scaling Language-Image Pre-training | openalex | 2022 | candidate |
Next steps
- Edit
m2d/config/experiment_brief.mdto refine the PICO criteria for screening. - Run
mine2digest next-candidates --agent-id A --limit 10to start the first screening pass. - Provide your
UNPAYWALL_EMAILin the.envfile to enable PDF fetching for confirmed papers.
Connects securely to your tools. The creator never sees your data.
About this skill
Researching literature for academic papers or technical reports often results in fragmented notes, lost citations, and biased selection. Manually verifying every claim against PDF text is slow and prone to "hallucinated" evidence.
This skill automates the literature pipeline from discovery to evidence extraction. It enforces a strict dual-agent screening process to eliminate bias, where two independent agents vote on candidates based on your specific research brief. It fetches full-text PDFs, extracts evidence cards with mandatory word-for-word source verification, and generates coverage matrices to identify gaps in your research. Every claim is back-linked to a specific text span in the source PDF using fuzzy locator matching, ensuring an auditable trail from draft to source.
What it does
- Dual-Agent Screening executes independent PICO and rule-based voting to filter candidates without cross-contamination.
- Evidence Extraction creates cards containing verbatim text spans from PDFs with automatic locator verification.
- Audit Logging maintains a SQLite database as the single source of truth for every paper state and evidence claim.
- Coverage Analysis generates matrices of sections versus evidence types to highlight research gaps.
- Draft Verification validates that every citation in a generated draft exists in the evidence database.
- Figure Analysis extracts figure captions and context to suggest design directions based on existing literature.
How it works
- Initialize the workspace using
mine2digest initto set up the local database and research brief. - Discover candidates by running seed scripts for OpenAlex or Semantic Scholar to populate the inbox.
- Screen papers by dispatching independent sub-tasks for Agent A and Agent B, then talling votes.
- Extract evidence by fetching PDFs and running the reader agent to generate verified evidence cards.
- Verify outputs by running coverage reports and checking drafts against the evidence database.
Frameworks & tools
Built for Python environments using SQLite for data persistence. It integrates with OpenAlex and Semantic Scholar APIs for discovery and Unpaywall for PDF retrieval. It uses rapidfuzz for text span verification and GitHub-flavored Markdown for all exports.
Why this beats prompting it yourself
Standard prompting often ignores source constraints or mixes up citations across multiple papers. This skill enforces a structured state machine and verbatim text matching that prevents the agent from "inventing" supporting evidence. The dual-blind screening logic ensures a higher level of systematic rigor than a single-pass prompt could achieve.
Use cases
- Conducting systematic literature reviews for academic publications.
- Building a verified knowledge base for technical whitepapers.
- Auditing existing drafts to ensure every claim has a direct source quote.
- Identifying missing evidence types in a specific section of a manuscript.
Known limitations
Requires external API keys for OpenAlex or Unpaywall for full functionality. Does not support direct PDF editing or complex multi-step reasoning outside the defined pipeline phases.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires UNPAYWALL_EMAIL in .env for PDF fetching. Database modification only via CLI. Evidence spans must be verbatim. Vision capabilities required for figure design analysis.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 3 days ago
- Free with an Agensi account