Open-source repos
Scientific Agent Skills
Appraise scientific evidence and prioritize traceable research directions
This page provides source information and editorial guidance. Recommendations and trust labels do not mean every skill here was tested. Results for individual skills are on the skill data pages. See skill data · Listing policy
Overview
This selection from K-Dense covers two research-method skills: scientific-critical-thinking examines claims, study design and evidence limits, while scientific-brainstorming supports independent idea generation, explicit criteria and decision records. scientific-schematics is preserved as optional diagram support. This page analyzes that selection without implying the entire scientific toolkit was collected.
SkillFlux sample check
On 2026-10-31, SkillFlux pinned this source at commit 5f2bc03, drew a seeded sample of 20 from the 177 entries it contains, and ran static health check 1.0.0 on each. 65% had no failing check, said when to use the skill and carried a license.
- 20/20sampled entries had a SKILL.md we could check
- 0/20fail at least one check
- 7/20do not say when to use the skill
- 0/20have no license we could find
- 17/20bundle scripts the agent may run
- 3/20include scripts that call the network
- 0/20include destructive shell commands
- 3/20exceed the 5,000-token body guideline
See the 20 sampled entries
| Entry | Result | Flags |
|---|---|---|
| K-Dense-AI/scientific-agent-skills/skills/scientific-schematics | 15 pass, 0 warn, 0 fail | Bundles scripts the agent may run, Scripts call the network |
| K-Dense-AI/scientific-agent-skills/skills/lab-hardware-cad | 14 pass, 1 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/scvelo | 15 pass, 0 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/scientific-slides | 13 pass, 2 warn, 0 fail | Bundles scripts the agent may run, Scripts call the network |
| K-Dense-AI/scientific-agent-skills/skills/peer-review | 14 pass, 1 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/optimize-for-gpu | 15 pass, 0 warn, 0 fail | — |
| K-Dense-AI/scientific-agent-skills/skills/nmrglue | 15 pass, 0 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/scvi-tools | 15 pass, 0 warn, 0 fail | — |
| K-Dense-AI/scientific-agent-skills/skills/scholar-evaluation | 15 pass, 0 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/treatment-plans | 15 pass, 0 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/torchdrug | 15 pass, 0 warn, 0 fail | — |
| K-Dense-AI/scientific-agent-skills/skills/venue-templates | 14 pass, 1 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/pydicom | 14 pass, 1 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/pymatgen | 15 pass, 0 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/shap | 15 pass, 0 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/bioservices | 14 pass, 1 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/paper-lookup | 14 pass, 1 warn, 0 fail | Bundles scripts the agent may run, Scripts call the network |
| K-Dense-AI/scientific-agent-skills/skills/scanpy | 14 pass, 1 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/analytical-method-validation | 15 pass, 0 warn, 0 fail | Bundles scripts the agent may run |
| K-Dense-AI/scientific-agent-skills/skills/pymoo | 14 pass, 1 warn, 0 fail | Bundles scripts the agent may run |
The sample reads files only; it does not run any skill and does not describe entries outside the sample. Flags describe what a skill contains; they are not verdicts. Raw data · Method
What the skills help you do
Inputs, methods and outputs described in the collected source version.
Extract evidence before judging a claim
Start with the question, sampling unit, results, denominators and uncertainty. Check whether design and analysis support the claim, while distinguishing missing reporting from a procedure not performed.
Read the source instructions ↗Write a traceable critique
Organize concerns by severity with a source location, observation, effect on the claim and feasible remedy. Treat reporting completeness, risk of bias and evidence certainty separately.
Read the source instructions ↗Preserve reasoning and dissent in ideation
Generate ideas independently before clustering, defining criteria, adversarial review and literature checks. Retain raw ratings, assumptions, dissent and the final decision rationale.
Read the source instructions ↗Optionally draft scientific diagrams
scientific-schematics accepts source-backed diagram descriptions and saves PNG drafts with review logs. Model scores help guide revision rather than establish journal acceptance.
Read the source instructions ↗A typical workflow
Choose appraisal or ideation
Provide a claim and its materials for appraisal or a focused question for ideation, recording goals, known information and unknowns.
Preserve evidence and idea provenance
For appraisal, extract results with their locations. For ideation, generate independently and assign stable IDs with assumptions, predictions and provenance.
Apply the appropriate evaluation process
Choose appraisal frameworks for the study design or define ideation criteria in advance, then examine alternative explanations, disagreement and evidence gaps.
Record a bounded next step
Deliver a traceable critique or decision log with remaining searches, validation and expert review tasks. Generate diagrams separately when needed.
Best for
- Researchers who need to locate evidence behind a paper or scientific claim and distinguish risk of bias from certainty of evidence.
- Research teams choosing directions who need to preserve dissent, evaluation reasons and follow-up validation tasks.
Use cases
- Coding & development
- Content & writing
- Planning & long tasks
SkillFlux notes
The analysis covers the two main skills and the scientific-schematics support skill, based on their instructions, references, scripts and MIT licenses at a pinned snapshot, without behavioral testing. It is not a scientific finding or an execution test; Python packages and external APIs must be prepared separately.
Installation and use
Open the original repository and read its README, the selected skill instructions and any linked scripts. Check installation paths, host requirements, dependencies and licenses, then follow the project documentation to verify a small task in a workspace you can restore.
Limitations
- Core analytical guidance works offline. Optional brainstorming utilities need Python 3.11+ and check structure or scoring records, not scientific truth.
- Diagram support needs Python, requests, an OpenRouter key and network access, and sends content to an external service. It produces PNG drafts whose labels and relationships need human checking; idea scores do not replace validation or expert review.
Alternatives
- Auto-Empirical Research Skills (AERS)
Compare the selected AERS workflows when the task concerns empirical-economics identification, analysis planning and replication packages.
- anthropics/skills
Compare Anthropic’s document and presentation skills when research content is settled and the main need is preparing deliverables.
Source links
- github.com/K-Dense-AI/scientific-agent-skills
- github.com/K-Dense-AI/scientific-agent-skills/blob/92ace75ac21efe19a620434e0ca4e356081fe807/README.md
- github.com/K-Dense-AI/scientific-agent-skills/blob/92ace75ac21efe19a620434e0ca4e356081fe807/LICENSE.md
- github.com/K-Dense-AI/scientific-agent-skills/blob/92ace75ac21efe19a620434e0ca4e356081fe807/docs/skills.md