Hypothesis generate assumptions
Skill panjose/Co-Scientist/skills/hypothesis-generate-assumptions
Scientific agent skills for Claude Code and Codex that turn research goals into auditable hypothesis generation, review, ranking, evolution, and synthesis.
npx -y skills add panjose/Co-Scientist --skill hypothesis-generate-assumptionsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Generate exactly one hypothesis candidate by enumerating and combining testable assumptions.
SKILL.md
3.6 KB, as published. Nobody here has run it
hypothesis-generate-assumptions
Goal:
- Generate exactly one hypothesis candidate by enumerating and combining testable assumptions.
Inputs:
research_plan/RESEARCH_PLAN.jsonstate/STRATEGY_PLAN.json- optional parent hypothesis and review artifacts when the round is part of an evolution continuation
Outputs:
hypotheses/<id>/HYPOTHESIS.jsonhypotheses/<id>/HYPOTHESIS.mdhypotheses/<id>/ORIGIN.json
Context Loading:
- Read
research_plan/RESEARCH_PLAN.json. - Use
research_goalas the objective that the assumption chain must explain or enable. - Use
preferencesas quality criteria. - Use
constraintsas hard boundaries. - Read
state/STRATEGY_PLAN.jsonand confirm that the current round allowsassumptions_identification_generation. - If the round is parented, read the selected parent hypothesis and its latest review summary before proposing a child. The new chain should address known weaknesses where possible.
Execution Prompt Contract:
- System Intent:
- You are generating one candidate hypothesis by surfacing the smallest useful chain of testable assumptions.
- Required Reasoning Focus:
- Identify 3-5 assumptions or fewer if a shorter chain is stronger.
- Favor chains that are falsifiable, mechanistically informative, and non-trivial.
- At least one link may be speculative, but it must remain testable and explicit.
- Use the assumption chain to produce a full downstream hypothesis rather than stopping at the decomposition.
- Do Not Do:
- Do not output multiple competing chains as final answers.
- Do not hide speculative links behind broad claims.
- Do not emit assumptions without turning them into a full canonical hypothesis artifact.
- Output Shape:
- The result must contain the exact
HypothesisContractfrompackages/agent_contracts/hypothesis.py. - If assumption structure is useful, keep it inside
origin-level payloads or trace notes, not as a replacement for the canonical hypothesis. origin.content.statement: 2-3 sentences maximum.origin.content.mechanism: 2-3 sentences maximum.origin.content.experimental_design: one concise multiline string with 3-6 numbered steps.origin.content.experimental_designmust remain one string field containing embedded line breaks; do not emit it as a list, array, or nested object.origin.content.summary: one sentence.origin.content.category: 1-5 words.
- The result must contain the exact
Execution Steps:
- Open
skills/shared-references/schema-index.md, then readpackages/agent_contracts/hypothesis.pyand confirm the exactHypothesisContractshape before writinghypotheses/<id>/HYPOTHESIS.json. - Read the required artifacts.
- Confirm that this round is allowed to use assumptions-driven generation.
- Identify the smallest useful chain of testable assumptions for the active goal.
- Synthesize the chain into exactly one candidate hypothesis.
- Wrap the result into the canonical
HypothesisContract. - Write
hypotheses/<id>/HYPOTHESIS.json,hypotheses/<id>/HYPOTHESIS.md, andhypotheses/<id>/ORIGIN.json. - Validate the emitted artifacts before declaring success.
Artifact Rules:
- The canonical hypothesis artifact is mandatory.
- Any auxiliary assumption tree must be treated as support for
origin, not as a substitute forHYPOTHESIS.json. - The final hypothesis must remain understandable even if a downstream consumer only reads the canonical artifact.
Completion Rule:
- This skill is complete only when exactly one new valid canonical hypothesis artifact has been written for the current round.