Hypothesis Generation
spacering-net/codeg
Structured hypothesis formulation from observations. An agent skill from spacering-net/codeg.
Two-perspective co-reasoning for AI4Science hypothesis generation.
Install Deltasci
$ npx skills add boheling/deltasci --skill deltasci -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install boheling/deltasci deltasci --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/boheling/deltasci.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skill .claude/skills/deltasci && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "deltasci" agent skill from https://github.com/boheling/deltasci/tree/main/skill into .claude/skills/deltasci/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "deltasci", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/boheling/deltasci/tree/main/skillType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add boheling/deltasci --skill deltasci -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install boheling/deltasci deltasci --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/boheling/deltasci.git skills-src && mkdir -p .agents/skills && cp -r skills-src/skill .agents/skills/deltasci && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "deltasci" agent skill from https://github.com/boheling/deltasci/tree/main/skill into .agents/skills/deltasci/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "deltasci", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add boheling/deltasci --skill deltasci -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install boheling/deltasci deltasci --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/boheling/deltasci.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/skill .cursor/skills/deltasci && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "deltasci" agent skill from https://github.com/boheling/deltasci/tree/main/skill into .cursor/skills/deltasci/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "deltasci", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/boheling/deltasci.git --path skill--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add boheling/deltasci --skill deltasci -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install boheling/deltasci deltasci --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/boheling/deltasci.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/skill .gemini/skills/deltasci && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "deltasci" agent skill from https://github.com/boheling/deltasci/tree/main/skill into .gemini/skills/deltasci/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "deltasci", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install boheling/deltasci deltasciInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add boheling/deltasci --skill deltasci -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/boheling/deltasci.git skills-src && mkdir -p .github/skills && cp -r skills-src/skill .github/skills/deltasci && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "deltasci" agent skill from https://github.com/boheling/deltasci/tree/main/skill into .github/skills/deltasci/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "deltasci", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add boheling/deltasci --skill deltasci -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install boheling/deltasci deltasci --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/boheling/deltasci.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/skill .opencode/skills/deltasci && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "deltasci" agent skill from https://github.com/boheling/deltasci/tree/main/skill into .opencode/skills/deltasci/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "deltasci", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
Facts
- Skill name
deltasci- GitHub stars
- 144
- Token cost
- ~1.8k tokens
- SKILL.md length
- 713 words
- Files
- 8 (incl. references)
- Skills in repo
- 2
- Repo updated
- First seen
- Licence
- MIT
At a glance
Two-perspective co-reasoning for AI4Science hypothesis generation.
- Works in 6 steps: Identify the domain pack → Run the 4-round dialogue → Write the transcript → …
- A researcher has a vague idea and wants to turn it into a defensible
- SKILL.md covers Purpose, The three first-class tags, Inputs and Outputs, plus 4 more sections
- Runs Shell scripts from its folder
What it does
Deltasci is an agent skill from boheling/deltasci. Two-perspective co-reasoning for AI4Science hypothesis generation. Runs a structured 4-round dialogue between a domain scientist (parameterized by a domain pack) and an ML engineer, producing a grounded, falsifiable research hypothesis that is honest about the AI's training-distribution edges. Domain-agnostic via pluggable packs (biomed, materials, climate, or your own). Use when a researcher has a vague idea and wants to turn it into a defensible, evaluable hypothesis with explicit handoffs for the things only…
Its SKILL.md is about 1.8k tokens, which your agent loads only when the skill is triggered. The skill folder holds 9 other files, including reference files (for example `install.sh`, `prompts/domain_round.md` and `prompts/engineer_round.md`).
It sits in Research & Science, covering Hypothesis generation. The repository describes itself as: A verification layer for scientific work. The licence is MIT.
When your agent uses it
- A researcher has a vague idea and wants to turn it into a defensible
- Evaluable hypothesis with explicit handoffs for the things only the researcher can know
Example prompts
- “/deltasci”
Requirements
- Python 3
- A Bash shell
Workflow steps
6 steps, taken from the step headings in SKILL.md.
What it can do on your machine
Read from SKILL.md and the folder at commit 5b36015. It shows what the files ask for, not the result of running them.
Tool permissions
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Runs code
Ships script files (Shell), which the agent can run.
From the folder's file list and the shell code blocks in SKILL.md.
Network
No URLs in SKILL.md.
From URLs in SKILL.md, links to its own repository left out.
Credentials
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Context cost
Deltasci loads about 1.8k tokens when it runs, and up to ~4.2k if it reads all its reference files. Until then it costs about 138 tokens; SKILL.md has 713 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
Safety
Auto-check passed
The automated check found no risky patterns in SKILL.md.
Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.
SKILL.md
The full file from boheling/deltasci at commit 5b36015, republished under its MIT licence (© boheling). 713 words, ~1,768 tokens.
Download SKILL.mdSave it as
.claude/skills/deltasci/SKILL.md (or your agent's skills folder). This skill also uses 7 other files; get the full folder from GitHub.- name
- deltasci
- description
- Two-perspective co-reasoning for AI4Science hypothesis generation. Runs a structured 4-round dialogue between a domain scientist (parameterized by a domain pack) and an ML engineer, producing a grounded, falsifiable research hypothesis that is honest about the AI's training-distribution edges. Domain-agnostic via pluggable packs (biomed, materials, climate, or your own). Use when a researcher has a vague idea and wants to turn it into a defensible, evaluable hypothesis with explicit handoffs for the things only the researcher can know.
DeltaScience: Two-Perspective Co-Reasoning
Purpose
Generate AI4Science research hypotheses that are:
- Grounded — every factual claim is tagged with its evidence type, source, and the AI's self-assessed training coverage.
- Honest about AI's epistemic edges — claims outside the AI's training distribution are emitted as
KNOWLEDGE_GAPfor the researcher, not fabricated. - Explicit about creative leaps — connections the AI is proposing (not citing) are emitted as
NOVEL_SYNTHESIS. - Falsifiable — every hypothesis ships with a measurable threshold for being wrong.
- Domain-aware — uses a pluggable lens specific to the scientific field.
- Two-perspective — alternates between domain expert and ML engineer.
This is the Claude Code skill version of the deltasci Python package. They share the same domain packs and grounding rules.
The three first-class tags
[CLAIM type=<TYPE> coverage=<COVERAGE> source="<CITATION>"]<text>[/CLAIM]
[KNOWLEDGE_GAP category=<CATEGORY>]<question for the researcher>[/KNOWLEDGE_GAP]
[NOVEL_SYNTHESIS rationale="<one-line>"]<the proposed connection>[/NOVEL_SYNTHESIS]coverage ∈ {well-covered, sparse}. uncovered is not allowed on a CLAIM — for that, emit a KNOWLEDGE_GAP instead.
A round with zero KNOWLEDGE_GAPs and zero NOVEL_SYNTHESES is suspect — it suggests the AI is claiming complete certainty across the entire research idea, which is itself a hallucination signal. Synthesis refuses by default.
Inputs
| Parameter | Required | Description |
|---|---|---|
idea | Yes | The raw research idea. |
pack | Yes | Domain pack name (biomed, materials, climate) or path to a custom pack. |
context_dir | No | Directory of background papers/notes. |
out_dir | No | Where to write outputs. Default: ./deltasci-output/. |
Outputs
| File | Description |
|---|---|
transcript.md | Full 4-round dialogue with all three tag types. |
hypothesis.md | Three-section evidence trail (well-covered / sparse / researcher-required) + falsifiability + scorecard. |
summary.json | Machine-readable hypothesis schema + epistemic summary. |
Step-by-Step Procedure
Step 1 — Identify the domain pack
Read <pack_dir>/pack.toml and <pack_dir>/lens.md. If the user's request doesn't name a pack and the domain is unambiguous, pick one without asking.
Step 2 — Run the 4-round dialogue
Follow the prompts in prompts/. The flow is:
Round 1 Domain Scientist (mechanism, unmet need, prior work, constraints)
Round 2 ML Engineer (data representation, method, precedents, risks)
Round 3 Domain Scientist (refinement, evaluation realism, falsifiable prediction)
Round 4 ML Engineer (revised plan, math, implementation, expected outcomes)In every round, every factual statement must be one of the three tags. See references/grounding_rules.md and references/coverage_axis.md for the rationale.
After each round, scan the output. If you find untagged factual claims, redo that round (one repair attempt). If a round has zero KNOWLEDGE_GAPs and zero NOVEL_SYNTHESES, prompt yourself: "Am I being honest about my training-distribution edges? Am I being honest about which connections are leaps vs cited?" — and rewrite if needed.
Step 3 — Write the transcript
Write to <out_dir>/transcript.md with each round labeled and all three tag types preserved.
Step 4 — Synthesize the grounded hypothesis
Follow prompts/synthesis_round.md. Produce JSON with:
title,statementdomain_grounding,technical_approachfalsifiability(prediction + threshold + null_outcome — all required)feasibility_scores,feasibility_justifications
The CLAIMs, KNOWLEDGE_GAPs, and NOVEL_SYNTHESES are collected automatically from the transcript — do NOT re-emit them in the synthesis JSON.
Hard rules:
- No falsifiable threshold → refuse, output
{"error": "no_falsifiable_clause", "reason": "..."}. - Across all rounds, zero KNOWLEDGE_GAPs and zero NOVEL_SYNTHESES → refuse, output
{"error": "no_epistemic_humility", "reason": "..."}.
Show full SKILL.md (279 more words)Show less
Step 5 — Render the hypothesis
The final hypothesis.md has THREE evidence sections, in this order:
- AI-confident foundations (well-covered claims with citations)
- Likely-reliable, please verify (sparse-coverage claims — cite carefully)
- Researcher knowledge required — KNOWLEDGE_GAPs + NOVEL_SYNTHESES
Plus an Epistemic summary with counts of each tag type and any warnings (e.g., "sparse claims outnumber well-covered ones").
Step 6 — Report
Tell the user:
- The hypothesis title
- Overall feasibility score
- Counts: well-covered / sparse / knowledge gaps / novel syntheses
- The falsifiability threshold
- Any epistemic warnings
- Where output files are
Configuration
| Setting | Default | Description |
|---|---|---|
num_rounds | 4 | Total dialogue rounds (must be even). |
grounding_strictness | high | Reject tagless claims with one repair attempt. |
require_falsifiability | true | Refuse to emit a hypothesis without a falsifiability clause. |
require_epistemic_humility | true | Refuse to emit a hypothesis when no KNOWLEDGE_GAPs and no NOVEL_SYNTHESES were emitted across the transcript. |
Rules
- Never invent citations. If you can't name a real reference, either tag the claim
coverage=sparseand use the most-honest source you have, or emit a KNOWLEDGE_GAP. - Never use
coverage=uncoveredon a CLAIM. Uncovered material goes in a KNOWLEDGE_GAP. - Mark your leaps as NOVEL_SYNTHESIS. Combining two well-covered facts into a hypothesis nobody has written down is good — it's the point. But say so explicitly.
- Skip the falsifiability gate is not allowed. A hypothesis without a measurable threshold is a wish.
- The pack lens is read literally. It parameterizes how the domain expert reasons.
Checklist
- Domain pack identified and lens loaded.
- 4 rounds completed with proper CLAIM / KNOWLEDGE_GAP / NOVEL_SYNTHESIS tags.
- At least one KNOWLEDGE_GAP and at least one NOVEL_SYNTHESIS across the transcript.
- Synthesis produced valid JSON matching the schema.
- Falsifiability clause has prediction + threshold + null_outcome.
- Hypothesis.md has all three evidence sections.
- Outputs written: transcript.md, hypothesis.md, summary.json.
© boheling, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
Files
SKILL.md and 7 other files (references) in skill of boheling/deltasci.
- SKILL.md
- install.sh
- prompts/domain_round.md
- prompts/engineer_round.md
- prompts/synthesis_round.md
- references/coverage_axis.md
- references/evidence_taxonomy.md
- references/grounding_rules.md
Open the folder on GitHubat commit 5b36015
Compare with similar skills
Deltasci next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Deltasci this skillboheling/deltasci | 144 | — | ~1.8k | Automated safety check: Pass | MIT | |
| Hypothesis Generationspacering-net/codeg | 3.9k | 14 repos | ~3.6k | Automated safety check: Notes | MIT | |
| Nature Paper CardYuan1z0825/nature-skills | 47k | 2 repos | ~2.1k | Automated safety check: Pass | Apache-2.0 | |
| Hypothesis GenerationK-Dense-AI/claude-scientific-writer | 2.4k | 2 repos | ~3.9k | Automated safety check: Pass | MIT | |
| Good QuestionRimagination/good-question | 305 | 1 repos | ~4.3k | Automated safety check: Pass | MIT | |
| High Stakes Analytics Decision Lablimingrui679-design/high-stakes-analytics-decision-lab | 1k | — | ~2.2k | Automated safety check: Pass | MIT |
Similar skills
Nature Paper Card
Yuan1z0825/nature-skills
Builds a structured deep-reading card for one scientific paper, covering methods, how experiments support claims, limitations and research ideas, with a script to prepare the source.
Hypothesis Generation
K-Dense-AI/claude-scientific-writer
Formulate evidence-bounded scientific questions, candidate hypotheses, rival explanations, causal or associational claims, discriminating predictions, measurements, and preregistration-ready…
Good Question
Rimagination/good-question
A skill your agent uses when a researcher is choosing, framing, refining, or stress-testing a research question, hypothesis, thesis topic, project idea, grant direction, paper angle, or stalled…
High Stakes Analytics Decision Lab
limingrui679-design/high-stakes-analytics-decision-lab
Build or review source-backed descriptive, diagnostic, predictive, and prescriptive analysis for consequential decisions.
Claim-Driven Experiment Planner
zjYao36/Auto-Research-Refine
Turns a refined research proposal into a claim-to-evidence-to-run-order roadmap instead of a sprawling benchmark wishlist.
More from boheling/deltasci
Deltasci Ground
boheling/deltasci
The grounding layer for AI-assisted research: scan the real prior art around an idea, gauge how crowded or open the gap is, and verify every citation against the source of truth.
Related
Categories
Questions about Deltasci
What does Deltasci do?
Two-perspective co-reasoning for AI4Science hypothesis generation. Deltasci is an agent skill from boheling/deltasci. Two-perspective co-reasoning for AI4Science hypothesis generation.
When should I use Deltasci?
Deltasci fits situations like: A researcher has a vague idea and wants to turn it into a defensible; evaluable hypothesis with explicit handoffs for the things only the researcher can know.
How do I install Deltasci in Claude Code?
Run `npx skills add boheling/deltasci --skill deltasci -a claude-code`. Or copy the skill folder (skill in boheling/deltasci) into .claude/skills/deltasci in your project. Claude Code loads it when a task matches its description.
How do I install Deltasci in Codex?
Run `npx skills add boheling/deltasci --skill deltasci -a codex`. Or copy the skill folder (skill in boheling/deltasci) into .agents/skills/deltasci in your project. Codex loads it when a task matches its description.
Can I use Deltasci in Cursor, Gemini CLI or GitHub Copilot?
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add boheling/deltasci --skill deltasci -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/deltasci, .gemini/skills/deltasci, .github/skills/deltasci and .opencode/skills/deltasci in your project.
What does Deltasci need to run?
Going by SKILL.md and its folder, Deltasci needs a shell for the scripts in its folder. Our summary lists: Python 3; A Bash shell.
Does Deltasci access the network?
SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.
Is Deltasci safe to install?
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.
What licence does Deltasci use?
Deltasci is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.
How many tokens does Deltasci use?
About 1.8k tokens (SKILL.md is roughly 7.1k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 2.4k tokens, read only when the agent opens those files.
What are the alternatives to Deltasci?
Skills that share tags, products or a category with Deltasci: Hypothesis Generation (spacering-net/codeg, 3.9k stars), Nature Paper Card (Yuan1z0825/nature-skills, 47k stars), Hypothesis Generation (K-Dense-AI/claude-scientific-writer, 2.4k stars) and Good Question (Rimagination/good-question, 305 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
Who maintains Deltasci?
boheling (a GitHub user) maintains it in boheling/deltasci, which has 144 GitHub stars. The repository holds 2 skills in this directory. The repository was last updated on May 30, 2026.
Source: boheling/deltasci on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.