Skip to content
Back to search

evoskills

๐Ÿงฌ Extend EvoScientist with Installable Skill & Knowledge Packs

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/evoscientist/evoskills with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/evoscientist/evoskills'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/evoscientist--evoskills/badge.svg)](/listings/evoscientist--evoskills)

HTML

<a href="/listings/evoscientist--evoskills"><img src="/api/listings/evoscientist--evoskills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

> [!IMPORTANT] > These skills are purpose-built for EvoScientist โ€” together they amplify each other, unlocking the full potential of both the agent and the skills. Under EvoScientist, skills evolve across research cycles through persistent memory (evo-memory).

In-session commands

Install all skills at once:

/install-skill EvoScientist/EvoSkills@skills

Or install a single skill:

/install-skill EvoScientist/EvoSkills@skills/paper-planning

Ask EvoScientist directly

Simply ask the agent in conversation:

"Install all skills from EvoScientist/EvoSkills@skills."

> [!TIP] > **Not using EvoScientist?** These skills are compatible with any coding agent. > One command via [**skills.sh**](https://skills.sh/) to install on Claude Code, OpenCode, Cursor, Codex, Gemini CLI, DeepAgents, and more: > ```bash > npx skills add EvoScientist/EvoSkills > ```

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md ยท README ยท 399 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # ๐Ÿงฌ EvoSkills
  2. <div align="center">
  3. **English | [็ฎ€ไฝ“ไธญๆ–‡](./README.zh-CN.md)**
  4. </div>
  5. **The official skill repository for [EvoScientist](https://github.com/EvoScientist/EvoScientist). Each skill is an installable knowledge pack that extends EvoScientist with domain-specific expertise.**
  6. ## ๐Ÿ“ฆ Installation
  7. > [!IMPORTANT]
  8. > These skills are purpose-built for EvoScientist โ€” together they amplify each other, unlocking the full potential of both the agent and the skills. Under EvoScientist, skills evolve across research cycles through persistent memory (evo-memory).
  9. ### In-session commands
  10. Install all skills at once:
  11. ```bash
  12. /install-skill EvoScientist/EvoSkills@skills
  13. ```
  14. Or install a single skill:
  15. ```bash
  16. /install-skill EvoScientist/EvoSkills@skills/paper-planning
  17. ```
  18. ### Ask EvoScientist directly
  19. Simply ask the agent in conversation:
  20. ```text
  21. "Install all skills from EvoScientist/EvoSkills@skills."
  22. ```
  23. > [!TIP]
  24. > **Not using EvoScientist?** These skills are compatible with any coding agent.
  25. > One command via [**skills.sh**](https://skills.sh/) to install on Claude Code, OpenCode, Cursor, Codex, Gemini CLI, DeepAgents, and more:
  26. > ```bash
  27. > npx skills add EvoScientist/EvoSkills
  28. > ```
  29. ## โœจ Available Skills
  30. | Skill | Description |
  31. | ----- | ----------- |
  32. | [`research-ideation`](#-research-ideation--literature-grounding-tournament--proposal) | ๐Ÿ’ก Literature grounding, tournament ranking & proposal generation |
  33. | [`paper-planning`](#-paper-planning--research-paper-planning--outline-generation) | ๐Ÿ“ Research paper planning & outline generation |
  34. | [`experiment-pipeline`](#-experiment-pipeline--4-stage-experiment-execution) | ๐Ÿงช Structured 4-stage experiment execution |
  35. | [`experiment-craft`](#-experiment-craft--experiment-debugging--iteration) | ๐Ÿ”ง Experiment debugging, logging & iteration |
  36. | [`experiment-iterative-coder`](#-experiment-iterative-coder--iterative-code-refinement) | ๐Ÿ”„ Iterative code refinement (plan โ†’ code โ†’ evaluate โ†’ refine) |
  37. | [`paper-writing`](#%EF%B8%8F-paper-writing--section-by-section-paper-drafting) | โœ๏ธ End-to-end paper writing assistance |
  38. | [`paper-review`](#-paper-review--self-review--quality-assurance) | ๐Ÿ” Automated paper review & feedback |
  39. | [`paper-rebuttal`](#-paper-rebuttal--rebuttal-writing-after-peer-review) | ๐Ÿ’ฌ Rebuttal writing after peer review |
  40. | [`paper-figures`](#-paper-figures--publication-ready-matplotlib-figures-from-data) | ๐Ÿ“Š Publication-ready matplotlib figures from tabular data |
  41. | [`academic-slides`](#-academic-slides--presentation--research-talk-creation) | ๐ŸŽค Academic presentation & research talk creation |
  42. | [`evo-memory`](#-evo-memory--persistent-research-memory--self-evolution) | ๐Ÿง  Persistent research memory & self-evolution |
  43. | [`paper-navigator`](#-paper-navigator--academic-paper-discovery--reading) | ๐Ÿ“š Academic paper discovery, evaluation & reading |
  44. | [`research-survey`](#-research-survey--literature-survey--synthesis) | ๐Ÿ“ Structured literature survey synthesis |
  45. | [`paper-graph`](#-paper-graph--lineage-map-of-a-research-field-via-mermaid-diagrams) | ๐ŸŒณ Lineage map of a research field as Mermaid diagrams |
  46. | [`nano-banana`](#-nano-banana--ai-generated-slides--illustrations) | ๐ŸŒ AI-generated presentation slides & illustrations via Gemini |
  47. | [`evomath-tao`](#-evomath-tao--tao-style-olympiad-proof-workflow) | ๐Ÿงฎ Tao-style olympiad-grade proof workflow with calibrated abstention |
  48. > **Paper Suite + Self-Evolution Suite**: Each skill is self-contained โ€” use them individually or combine freely. The self-evolution loop now runs through `research-ideation`, `experiment-pipeline`, and `evo-memory`.
  49. ## ๐Ÿ”Œ MCP Server Marketplace
  50. The [`mcp/`](./mcp/) directory contains a curated collection of [MCP](https://modelcontextprotocol.io/) servers that extend agents with external tools โ€” web search, academic paper retrieval, documentation lookup, and more. Browse the [full list](./mcp/README.md) or install directly:
  51. ```bash
  52. /install-mcp # interactive browser
  53. EvoSci mcp install arxiv # install by name
  54. ```
  55. ### โ›ณ๏ธ Framework Overview
  56. <p align="center">
  57. <img src="assets/framework.png" alt="EvoScientist Framework" width="100%" />
  58. </p>
  59. The diagram above shows the full EvoScientist pipeline. The **Researcher Agent** (top, blue) runs idea tree search and Elo tournament ranking to produce a research proposal (in this repository, the `research-ideation` skill: three persona-driven refinement tracks whose champions are ranked in a three-match Elo round-robin). The **Engineer Agent** (bottom, green) executes the 4-stage experiment pipeline. The **Evolution Manager Agent** (right) manages three memory evolution mechanisms โ€” IDE, IVE, and ESE โ€” that feed learned knowledge back into **Ideation Memory (M_I)** and **Experimentation Memory (M_E)** for future cycles.
  60. #### ๐ŸŽข Skill Pipeline
  61. ```mermaid
  62. flowchart LR
  63. A["<b>๐Ÿ”ฌ Research Phase</b><br/>๐Ÿ’ก research-ideation"]
  64. --> B["<b>โš™๏ธ Experiment Phase</b><br/>๐Ÿ“ paper-planning<br/>๐Ÿงช experiment-pipeline<br/>๐Ÿ”ง experiment-craft<br/>๐Ÿ”„ experiment-iterative-coder"]
  65. --> C["<b>๐Ÿ“ Writing Phase</b><br/>โœ๏ธ paper-writing<br/>๐Ÿ” paper-review<br/>๐Ÿ’ฌ paper-rebuttal<br/>๐ŸŽค academic-slides"]
  66. D[("๐Ÿง  evo-memory<br/>(IDE ยท IVE ยท ESE)")] <--> A
  67. D <--> B
  68. E["๐Ÿ“š paper-navigator<br/>(standalone)"] -.-> A
  69. E -.-> B
  70. F["๐ŸŒ nano-banana<br/>(standalone)"] -.-> C
  71. G["๐Ÿงฎ evomath-tao<br/>(standalone)"] -.-> A
  72. G -.-> B
  73. H["๐Ÿ“ research-survey<br/>(standalone)"] -.-> A
  74. H -.-> C
  75. I["๐ŸŒณ paper-graph<br/>(standalone)"] -.-> A
  76. J["๐Ÿ“Š paper-figures<br/>(standalone)"] -.-> C
  77. style A fill:#7C3AED,stroke:#5B21B6,stroke-width:2px,color:#fff
  78. style B fill:#D97706,stroke:#B45309,stroke-width:2px,color:#fff
  79. style C fill:#16A34A,stroke:#15803D,stroke-width:2px,color:#fff
  80. style D fill:#475569,stroke:#334155,stroke-width:2px,color:#fff
  81. style E fill:#0369A1,stroke:#075985,stroke-width:2px,color:#fff
  82. style F fill:#D97706,stroke:#B45309,stroke-width:2px,color:#fff
  83. style G fill:#BE185D,stroke:#9D174D,stroke-width:2px,color:#fff
  84. style H fill:#0891B2,stroke:#0E7490,stroke-width:2px,color:#fff
  85. style I fill:#65A30D,stroke:#4D7C0F,stroke-width:2px,color:#fff
  86. style J fill:#9333EA,stroke:#7E22CE,stroke-width:2px,color:#fff
  87. ```
  88. ---
  89. ### ๐Ÿ’ก `research-ideation` โ€” Literature Grounding, Tournament & Proposal
  90. The starting point of the research pipeline. It now covers the full path from literature grounding to ranked ideas to a concrete proposal:
  91. - **Load Prior Knowledge** โ€” Read `evo-memory` first to reuse feasible directions and avoid known dead ends
  92. - **Literature Grounding** โ€” Use `paper-navigator` to collect and analyze papers before generating ideas
  93. - **Multi-Track Ideation + Refinement** โ€” Generate one idea per research direction, then refine each in its own track with three personas (innovator, pragmatist, critic)
  94. - **Elo Tournament** โ€” Rank the three track champions on novelty, feasibility, relevance, and clarity; present all three, ranked
  95. - **Proposal Extension** โ€” Expand the selected winner into a manuscript-quality research proposal
  96. ### ๐Ÿ“ `research-survey` โ€” Literature Survey & Synthesis
  97. Dedicated skill for turning a large paper collection into a structured survey report:
  98. - **Adaptive Outline** โ€” Generate a field-specific outline based on the query type and literature set
  99. - **Draft + Expansion Pipeline** โ€” Draft from top papers, then deepen each section with the full collection
  100. - **Summary Refinement** โ€” Build section summaries before rewriting the abstract, introduction, and conclusion
  101. - **Survey-Grade Output** โ€” Comparative tables, taxonomy-based method organization, dense citations, and references
  102. ### ๐Ÿ“ `paper-planning` โ€” Research Paper Planning & Outline Generation
  103. Guides pre-writing planning before a single word is drafted. Covers four key activities:
  104. - **Story Design** โ€” Reverse-engineer the narrative: task โ†’ challenge โ†’ insight โ†’ contribution โ†’ advantage
  105. - **Experiment Planning** โ€” Plan comparisons, ablations, and demo scenarios with structured checklists
  106. - **Figure Design** โ€” Pipeline figures that highlight novelty; teaser figures that hook reviewers
  107. - **Timeline Management** โ€” 4-week countdown schedule from outline to submission
  108. Includes counterintuitive tactics: write your rejection letter first, narrow claims before broadening, and plan fallback narratives.
  109. ### ๐Ÿงช `experiment-pipeline` โ€” 4-Stage Experiment Execution
  110. A structured framework for executing research experiments with attempt budgets and gate conditions:
  111. - **Stage 1: Initial Implementation** โ€” Get baseline code running and reproduce known results (โ‰ค20 attempts)
  112. - **Stage 2: Hyperparameter Tuning** โ€” Optimize configuration for your setup (โ‰ค12 attempts)
  113. - **Stage 3: Proposed Method** โ€” Implement and validate the novel method (โ‰ค12 attempts)
  114. - **Stage 4: Ablation Study** โ€” Prove each component's contribution (โ‰ค18 attempts)
  115. - **Code Trajectory Logging** โ€” Structured attempt logging that feeds into `evo-memory`
  116. - **Counterintuitive Rules** โ€” Initial implementation is not wasted time; budget limits prevent rabbit holes; failed attempts are data
  117. Integrates with `experiment-craft` for failure diagnosis within stages and `evo-memory` for cross-cycle learning.
  118. ### ๐Ÿ”ง `experiment-craft` โ€” Experiment Debugging & Iteration
  119. A systematic approach to experiment debugging, logging, and iterative improvement:
  120. - **5-Step Diagnostic Flow** โ€” Collect failures โ†’ find a working version โ†’ bridge the gap โ†’ hypothesize โ†’ fix
  121. - **Counterintuitive Rules** โ€” Change one variable at a time; effective experiments beat more experiments
  122. - **Experiment Logging** โ€” 5-section structured log template for reproducible records
  123. - **Handoff to Paper-Writing** โ€” Feed validated results and logs into `paper-writing` for drafting
  124. ### ๐Ÿ”„ `experiment-iterative-coder` โ€” Iterative Code Refinement
  125. Structured plan โ†’ code โ†’ evaluate โ†’ refine cycles for higher code quality:
  126. - **Phase Decomposition** โ€” Break complex tasks into 1-5 sequential phases
  127. - **Iteration Loop** โ€” Up to 3 iterations per phase (10 total): plan, code, run lint/tests, score, decide
  128. - **Objective Evaluation** โ€” ruff lint + pytest with dynamic score weighting and hard caps
  129. - **Failure Mode Guidance** โ€” Targeted responses for timeout, syntax, import, test, and lint failures
  130. Integrates with `experiment-craft` for stuck diagnoses and `evo-memory` for loading prior strategies.
  131. ### โœ๏ธ `paper-writing` โ€” Section-by-Section Paper Drafting
  132. A proven 11-step workflow for writing academic papers with LaTeX templates:
  133. - **Structured Process** โ€” From pipeline sketch โ†’ story design โ†’ Method โ†’ Experiments โ†’ Related Work โ†’ Abstract โ†’ Title
  134. - **Section Templates** โ€” Three Abstract templates, four Introduction openers, Method module structure, Experiments organization
  135. - **LaTeX Assets** โ€” Annotated paper skeleton (`paper-skeleton.tex`) and booktabs table macros (`table-style.tex`)
  136. - **Writing Principles** โ€” One message per paragraph, topic sentence first, terminology consistency, reverse-outlining
  137. - **Counterintuitive Tactics** โ€” Underclaim in prose / overdeliver in evidence; lead with mechanism, not just metrics
  138. ### ๐Ÿ” `paper-review` โ€” Self-Review & Quality Assurance
  139. Systematic self-review before submission using adversarial and counterintuitive review strategies:
  140. - **5-Aspect Checklist** โ€” Contribution sufficiency, writing clarity, results quality, testing completeness, method design
  141. - **Reverse-Outlining** โ€” Extract the outline from finished paragraphs to verify logical flow
  142. - **Figure & Table Quality Checks** โ€” Captions, resolution, booktabs, color-blind friendliness
  143. - **Rejection Simulation** โ€” Force a reject summary first; attack your own novelty claim
  144. - **Handoff to Rebuttal** โ€” After review, feed identified weaknesses into `paper-rebuttal` for response preparation
  145. ### ๐Ÿ’ฌ `paper-rebuttal` โ€” Rebuttal Writing After Peer Review
  146. Dedicated rebuttal skill for responding to reviewer feedback after peer review:
  147. - **Score Diagnosis** โ€” Color-code every reviewer comment: red (critical), orange (important), gray (minor), green (positive)
  148. - **Champion Strategy** โ€” Arm your most positive reviewer with evidence for the Area Chair discussion
  149. - **Tactical Writing** โ€” 18 rules for structure, content, and tone in rebuttal responses
  150. - **Counterintuitive Principles** โ€” Submit even with extreme scores; concede small points to win the big argument
  151. - **Common Concerns** โ€” Response strategies for 12 frequently raised reviewer complaints
  152. ### ๐Ÿ“Š `paper-figures` โ€” Publication-Ready Matplotlib Figures from Data
  153. A spec-first workflow that turns CSVs and natural-language descriptions into standalone PNGs and reproducible matplotlib scripts:
  154. - **7-Step Protocol** โ€” Plan Figure โ†’ Inspect Data โ†’ Write `figure-spec.md` โ†’ Pick matplotlib idiom โ†’ Apply publication style โ†’ Write and run the script โ†’ Audit, with `plot.py` + `plot.png` as the always-paired output
  155. - **Spec-First Discipline** โ€” Every figure is preceded by a compact `figure-spec.md` contract (axes, scales, series, forbidden elements, assumptions) that the audit checks against
  156. - **4 Honest Status Labels** โ€” `PASSED` / `PASSED_WITH_WARNINGS` / `REPAIRED` / `FAILED_NEEDS_HANDOFF`. The script executing is not proof the figure matches the request
  157. - **Structural Audit, Not Visual** โ€” Verifies title, axis labels, series ordering, colors, annotations, and axis bounds against the description (LLM visual inspection of PNGs is unreliable)
  158. - **Broad Chart Coverage** โ€” Scatter, line, bar, pie, ring, bubble, tornado, KDE, violin, box, heatmap, histogram, area, and composite multi-panel figures
  159. - **Counterintuitive Rules** โ€” Don't drop data silently; don't invent uninvited elements; CSV is authoritative when it disagrees with the description; tight framing that crops a named feature is equivalent to silent data dropping
  160. ### ๐ŸŽค `academic-slides` โ€” Presentation & Research Talk Creation
  161. A structured approach to creating academic presentations and preparing research talks:
  162. - **Narrative Arc** โ€” Define scope, audience, and key takeaway before touching slides
  163. - **Slide Design** โ€” 10 design rules, visual hierarchy, one idea per slide, claim-style titles
  164. - **Practical Creation** โ€” `.pptx` file generation with color palettes, layout code, charts, and figures
  165. - **Delivery & Q&A** โ€” Rehearsal protocol, timing, and backup slide preparation
  166. - **Counterintuitive Rules** โ€” Slides are not your paper; enthusiasm beats polish; related work builds motivation, not citation counts
  167. ### ๐Ÿง  `evo-memory` โ€” Persistent Research Memory & Self-Evolution
  168. The learning layer that accumulates knowledge across research cycles. Maintains two memory stores and implements three evolution mechanisms:
  169. - **Ideation Memory (M_I)** โ€” Tracks feasible and unsuccessful research directions across ideation cycles
  170. - **Experimentation Memory (M_E)** โ€” Stores reusable data processing and model training strategies (paper core), plus architecture and debugging (extensions)
  171. - **IDE (Idea Direction Evolution)** โ€” Extracts promising directions after `research-ideation`
  172. - **IVE (Idea Validation Evolution)** โ€” Classifies experiment failures as implementation vs fundamental direction failures
  173. - **ESE (Experiment Strategy Evolution)** โ€” Distills reusable patterns from successful experiment pipelines
  174. Read by `research-ideation` and `experiment-pipeline` at cycle start; updated after each cycle completes.
  175. ### ๐Ÿ“š `paper-navigator` โ€” Academic Paper Discovery & Reading
  176. Finds and reads academic papers, routing by intent and judging by relevance:
  177. - **Router** โ€” A known paper (title, URL, or ID) takes the single-call POINT branch and returns a Paper Card; "find papers about X" and survey-scale requests take the LIST / ITERATIVE branch
  178. - **Rubric First** โ€” On the LIST / ITERATIVE branch, the agent writes a rubric of core (must-have) and secondary criteria before any search; every later step checks papers against it
  179. - **Probe + Up to 3 Rounds** โ€” A two-query probe, then up to three keyword-search rounds (breadth, deepen, close). With a Semantic Scholar API key, citation traversal and recommendations widen the search for survey-scale requests, or once at least three papers have passed triage. A saturation gate after the probe and after each round decides whether key gaps remain
  180. - **Triage Without Scores** โ€” Each paper is marked All-core / Partial / Irrelevant against the rubric, with a quoted span as evidence for every criterion it is said to meet; when an abstract is silent on a core criterion, body-text snippets can settle it (Semantic Scholar API key required). No LLM-as-judge, no numeric scoring
  181. - **Read** โ€” Full-text retrieval via Jina Reader with 3-level reading strategy (Technical, Analytical, Contextual)
  182. Includes Python scripts powered by Semantic Scholar, arXiv (via the DeepXiv SDK), HuggingFace, GitHub, and Jina Reader.
  183. ### ๐ŸŒณ `paper-graph` โ€” Lineage Map of a Research Field via Mermaid Diagrams
  184. A skill that turns a research topic or seed paper into a Markdown report tracing how a field evolved โ€” challenges, solutions, and per-solution citation lineage, all rendered as embedded Mermaid diagrams:
  185. - **Two Graph Layers** โ€” High-level taxonomy (root โ†’ challenges โ†’ solutions โ†’ papers) plus one per-solution evolution path tracing paper-to-paper "evolution from" edges and open challenges
  186. - **Agent-Driven LLM Calls** โ€” The skill ships deterministic data fetchers (Semantic Scholar / DeepXiv), prompt templates, and Mermaid renderers; the host agent runs every LLM stage, so the skill needs no model API key of its own (a Semantic Scholar API key is required for fetching papers)
  187. - **Edge Audit Pass** โ€” Every claimed "evolution from" edge is audited by a separate LLM step, then re-checked by the renderer: a verbatim quote from each paper must be found in its text and known publication years must be in order, or the edge is not drawn
  188. - **Renders Anywhere** โ€” Mermaid in fenced Markdown blocks views directly in GitHub, Obsidian, VS Code (with a Mermaid extension), and other Markdown viewers โ€” no external rendering pipeline
  189. - **Use Cases** โ€” "History of <topic>", "What did <paper> build on?", "Lineage of ideas in <field>", "Citation tree of <paper>"
  190. ### ๐ŸŒ `nano-banana` โ€” AI-Generated Slides & Illustrations
  191. Generate professional presentation slides and high-quality illustrations using Gemini's image generation API, with an interactive browser-based review loop:
  192. - **7-Phase Workflow** โ€” Content planning conversation โ†’ slides_plan.json โ†’ style selection & batch generation โ†’ browser review โ†’ feedback editing โ†’ PPTX packaging โ†’ cleanup
  193. - **3 Visual Styles** โ€” Lineal Color (flat icons, educational), Gradient Glass (glassmorphism, premium), Vector Illustration (retro, approachable)
  194. - **Interactive Review** โ€” Local HTTP server with per-slide feedback; edits are applied without regenerating the entire deck
  195. - **Multi-Model Support** โ€” `gemini-3-pro-image-preview` (best quality), `gemini-3.1-flash-image-preview` (fast iteration), `gemini-2.5-flash-image` (rapid prototyping)
  196. - **Counterintuitive Rules** โ€” More planning = better slides; edit don't regenerate; never read generated images yourself (use the review server)
  197. ### ๐Ÿงฎ `evomath-tao` โ€” Tao-style Olympiad Proof Workflow
  198. A rigorous proof workflow that operationalizes Terence Tao's research-math practice for contest-style mathematics. Outputs a complete proof, a verified counterexample, a calibrated partial result, or a clean handoff โ€” never a hand-waved "PROVED":
  199. - **5-Step Protocol** โ€” Plan Briefly โ†’ Try Candidates โ†’ Assemble โ†’ Audit โ†’ Reflect, driven by TodoWrite + per-phase validators
  200. - **5-Round Internal Mini-Process per Candidate** โ€” Solve โ†’ Self-improve โ†’ Self-verify โ†’ Correct โ†’ Repeat. **No tool use during Solve** (pencil-and-paper discipline)
  201. - **5 Honest Status Labels** โ€” `PROVED` / `REFUTED` / `VERIFIED_NUMERICALLY` / `CONJECTURED` / `HANDED_OFF`. Numerical evidence is NOT a proof
  202. - **3-Safeguard Audit** โ€” Verifier context isolation, asymmetric voting (4 HOLDS to confirm, 2 HOLE FOUND to refute), pigeonhole exit
  203. - **Named-Pattern Screen** โ€” Library of recurring failure modes (P4, P5, P6, P18, P40, P41) checked before any PROVED award
  204. - **Calibrated Abstention** โ€” When verification fails repeatedly, the skill downgrades the status instead of bluffing
  205. **IMO 2025 evaluation (Claude Opus 4.7, 6 parallel subagents):** 4 PROVED (P1 / P2 / P4 / P5) ยท 1 CONJECTURED (P3, c = 4 with odd-prime gap) ยท 1 HANDED_OFF (P6, 2112 sketched). All 6 numeric/classification answers match the official IMO 2025 keys.
  206. <p align="right"><a href="#top">๐Ÿ”Back to top</a></p>
  207. ## ๐ŸŽฏ แฏ“โžค Roadmap
  208. Completed:
  209. - [x] ๐Ÿง  **Self-Evolution Suite** โ€” `research-ideation`, `experiment-pipeline`, `evo-memory`
  210. - [x] ๐Ÿ“š **Literature Survey** โ€” Systematic literature search, filtering, and survey generation
  211. - [x] ๐Ÿ”„ **Iterative Coder** โ€” Iterative code refinement with plan โ†’ code โ†’ evaluate โ†’ refine cycles
  212. - [x] ๐ŸŽจ **Visual Generation** โ€” AI-generated slides & illustrations (`nano-banana`)
  213. - [x] ๐Ÿ… **Math Olympiad** โ€” Tao-style proof workflow with calibrated abstention (`evomath-tao`)
  214. Coming soon:
  215. - [ ] ๐Ÿ”ฌ **Paper Reproduction** โ€” Read a paper, reproduce its core results, and verify claims
  216. - [ ] ๐Ÿ’ก **Grant & Proposal Writing** โ€” Research proposal drafting with funding agency conventions
  217. - [ ] ๐Ÿค– **Peer Debate** โ€” Multi-agent adversarial discussion to stress-test research ideas
  218. - [ ] ๐Ÿ“ˆ **Trend Radar** โ€” Analyze publication trends, identify emerging topics & research gaps
  219. - [ ] ๐Ÿ—ฃ๏ธ **Paper QA** โ€” Interactive question-answering over paper collections, extracting key findings & cross-referencing claims
  220. Stay tuned โ€” more skills are on the way!
  221. <p align="right"><a href="#top">๐Ÿ”Back to top</a></p>
  222. ## ๐ŸŒ Project Roles
  223. <table>
  224. <tbody>
  225. <tr>
  226. <td align="center">
  227. <a href="https://x-izhang.github.io/">
  228. <img src="https://x-izhang.github.io/author/xi-zhang/avatar.jpg"
  229. width="100" height="100"
  230. style="object-fit: cover; border-radius: 20%;" alt="Xi Zhang"/>
  231. <br />
  232. <sub><b>Xi Zhang</b></sub>
  233. </a>
  234. </td>
  235. <td align="center">
  236. <a href="https://youganglyu.github.io/">
  237. <img src="https://youganglyu.github.io/images/profile.png"
  238. width="100" height="100"
  239. style="object-fit: cover; border-radius: 20%;" alt="Yougang Lyu"/>
  240. <br />
  241. <sub><b>Yougang Lyu</b></sub>
  242. </a>
  243. </td>
  244. <td align="center">
  245. <a href="https://din0s.me/">
  246. <img src="https://din0s.me/images/pk.jpg"
  247. width="100" height="100"
  248. style="object-fit: cover; border-radius: 20%;" alt="Dinos Papakostas"/>
  249. <br />
  250. <sub><b>Dinos Papakostas</b></sub>
  251. </a>
  252. </td>
  253. <td align="center">
  254. <a href="https://go0day.github.io/">
  255. <img src="https://go0day.github.io/authors/admin/avatar_hu_ee1051aceae96124.png"
  256. width="100" height="100"
  257. style="object-fit: cover; border-radius: 20%;" alt="Yuyue Zhao"/>
  258. <br />
  259. <sub><b>Yuyue Zhao</b></sub>
  260. </a>
  261. </td>
  262. </tr>
  263. </tbody>
  264. </table>
  265. > <a href="https://xiaoyi.huawei.com/chat/research"><img src="./assets/xiaoyi-deepresearch.webp" height="16" alt="Xiaoyi DeepResearch" align="center" /></a> [*Xiaoyi DeepResearch*](https://xiaoyi.huawei.com/chat/research) *Team* and the wider open-source community contribute to this project.
  266. For any enquiries or collaboration opportunities, please contact: [**EvoScientist.ai@gmail.com**](mailto:evoscientist.ai@gmail.com)
  267. <p align="right"><a href="#top">๐Ÿ”Back to top</a></p>
  268. ## ๐Ÿค Contributing
  269. We welcome contributions! See the guides for [skills](./skills/README.md) and [MCP servers](./mcp/README.md), or start with the [Contributing Guidelines](./CONTRIBUTING.md).
  270. Every contribution brings us one step closer to a future where AI accelerates scientific breakthroughs for all of humanity.
  271. <a href="https://github.com/EvoScientist/EvoSkills/graphs/contributors">
  272. <img src="https://contrib.rocks/image?repo=EvoScientist/EvoSkills" />
  273. </a>
  274. ### ๐Ÿ“ˆ Star History
  275. [![Star History Chart](https://api.star-history.com/chart?repos=EvoScientist/EvoSkills&type=date&legend=top-left&sealed_token=aiiKT_W8lO0VUktO2hVHL7jmLFkHrKF6SjzkDilktnITq2poApMnar2rs40ohkpGEPn2H_mKUfWXzhHxEzOPNS0cICz9SeItmovfPYWNojH3QqKE36CWvg)](https://www.star-history.com/?repos=EvoScientist%2FEvoSkills&type=date&legend=top-left)
  276. <p align="right"><a href="#top">๐Ÿ”Back to top</a></p>
  277. ## ๐Ÿ“ Citation
  278. If you find our paper and code useful in your research and applications, please cite using this BibTeX:
  279. ```bibtex
  280. @article{evoscientist2026,
  281. title={EvoScientist: Towards Multi-Agent Evolving AI Scientists for End-to-End Scientific Discovery},
  282. author={Yougang Lyu and Xi Zhang and Xinhao Yi and Yuyue Zhao and Shuyu Guo and Wenxiang Hu and Jan Piotrowski and Jakub Kaliski and Jacopo Urbani and Zaiqiao Meng and Lun Zhou and Xiaohui Yan},
  283. journal={arXiv preprint arXiv:2603.08127},
  284. year={2026}
  285. }
  286. ```
  287. <p align="right"><a href="#top">๐Ÿ”Back to top</a></p>
  288. ## ๐Ÿ“œ License
  289. This project is licensed under the Apache License 2.0 - see the [LICENSE](./LICENSE) file for details.
  290. <p align="right"><a href="#top">๐Ÿ”Back to top</a></p>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If itโ€™s helped, pass it on.