| name | hypothesis-evolve-grounding |
| description | Generate exactly one grounded child hypothesis by strengthening evidence, specificity, and literature support. |
hypothesis-evolve-grounding
Goal:
- Generate exactly one grounded child hypothesis by strengthening evidence, specificity, and literature support.
Inputs:
research_plan/RESEARCH_PLAN.json
- selected parent
hypotheses/<id>/HYPOTHESIS.json artifacts
- parent review artifacts
- active
state/STRATEGY_PLAN.json
Outputs:
hypotheses/<id>/HYPOTHESIS.json
hypotheses/<id>/HYPOTHESIS.md
hypotheses/<id>/ORIGIN.json
Context Loading:
- Open
skills/shared-references/schema-index.md.
- Read
packages/agent_contracts/hypothesis.py and confirm the exact HypothesisContract shape before writing hypotheses/<id>/HYPOTHESIS.json.
- Read
research_plan/RESEARCH_PLAN.json.
- Read
state/STRATEGY_PLAN.json.
- Read each selected parent hypothesis and its review bundle.
- Confirm that the round selected
grounding_evolution.
Execution Prompt Contract:
- System Intent:
- You are refining the parent hypothesis by replacing weakly supported claims with better-grounded, more specific, and more internally consistent ones.
- Required Reasoning Focus:
- Preserve the strongest core insight from the parent.
- Use review findings to identify unsupported leaps, vague mechanisms, or weak evidence.
- Strengthen the mechanism and experiment path without bloating the claim.
- Do Not Do:
- Do not merely paraphrase the parent.
- Do not emit multiple children.
- Do not drop the parent linkage.
- Quality Floor:
- The child must directly address at least one specific weakness from the parent review bundle.
origin.content.statement must name concrete materials, catalysts, reaction conditions, mechanistic variables, or experimental targets from the parent and research goal.
origin.content.mechanism must explain a causal chain; do not write only generic phrases such as improved mechanism, targeted improvement, or review-identified weaknesses.
origin.content.experimental_design must include 3-6 numbered steps with measurable readouts, controls, or decision thresholds.
- Do not use generic refinement placeholder steps such as
Apply targeted improvement, Characterize with standard techniques, Benchmark against parent, or Validate improvement quantitatively.
- If the research plan, parent hypothesis, or parent review bundle is missing, stop and report the missing artifact instead of guessing.
- Output Shape:
- Emit the canonical
HypothesisContract.
origin.strategy: grounding_evolution
origin.content.statement: 2-3 sentences.
origin.content.mechanism: 2-3 sentences.
origin.content.experimental_design: one concise multiline string with 3-6 numbered steps.
origin.content.experimental_design must 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.
Execution Steps:
- Open
skills/shared-references/schema-index.md, then read packages/agent_contracts/hypothesis.py before writing hypotheses/<id>/HYPOTHESIS.json.
- Identify the parent weaknesses that must be corrected by grounding.
- Produce exactly one improved child hypothesis.
- Persist canonical
HYPOTHESIS.json, HYPOTHESIS.md, and ORIGIN.json.
- Validate the emitted hypothesis artifact.
Completion Rule:
- This skill is complete only when one valid grounded child hypothesis has been written in canonical form.