Hypothesis observation review
Skill panjose/Co-Scientist/skills/hypothesis-observation-review
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-observation-reviewAssembled 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
Evaluate a hypothesis against prior observations.
SKILL.md
2.9 KB, as published. Nobody here has run it
hypothesis-observation-review
Goal:
- Evaluate a hypothesis against prior observations.
Inputs:
research_plan/RESEARCH_PLAN.jsonhypotheses/<id>/HYPOTHESIS.json- optional review guidance from prior review artifacts
Outputs:
hypotheses/<id>/REVIEW/OBSERVATION_REVIEW.json
Context Loading:
- Read
research_plan/RESEARCH_PLAN.json. - Read
hypotheses/<id>/HYPOTHESIS.json. - Use the current hypothesis mechanism and predictions as the reference point for comparing against known observations.
- If prior reviews exist, use them only to focus attention on disputed claims.
Execution Prompt Contract:
- System Intent:
- You are evaluating whether the hypothesis meaningfully explains relevant observations.
- Required Reasoning Focus:
- Look for observations that are causally relevant to the hypothesis domain.
- Judge whether the hypothesis explains them, adds nothing, is contradicted by them, or is less plausible than existing explanations.
- Prefer causal explanatory value over loose topical overlap.
- Do Not Do:
- Do not list observations that have no material connection to the hypothesis.
- Do not treat generic correlation as causal support.
- Do not emit free-form prose instead of structured observation entries.
- Review Quality Floor:
- A
status = completedobservation review must include concrete observations or retrieval results that are materially connected to the hypothesis mechanism or predicted behavior. - Each observation must explain why it supports, weakens, or remains neutral for the hypothesis rather than only naming a topic.
- Do not use placeholder observation phrases such as
consistent with prior observations,needs validation, orbenchmark against parentas substantive review content.
- A
- Output Shape:
- Produce the exact
ObservationReviewContractfrompackages/agent_contracts/review.py. - Each observation entry must contain:
reasoningconclusion
- Use the allowed conclusion categories only.
- Produce the exact
Execution Steps:
- Open
skills/shared-references/schema-index.md, then readpackages/agent_contracts/review.pyand confirm the exactObservationReviewContractshape before writingOBSERVATION_REVIEW.json. - Read the research plan and hypothesis.
- Identify the most relevant observations for the hypothesis domain.
- Evaluate whether the hypothesis explains, conflicts with, or adds little to each observation.
- Write
hypotheses/<id>/REVIEW/OBSERVATION_REVIEW.json. - Validate before declaring completion.
Artifact Rules:
- Limit the review to the strongest and most relevant observations rather than padding the artifact.
- Each observation entry must remain decision-oriented and concise.
Completion Rule:
- This skill is complete only when
OBSERVATION_REVIEW.jsonexists and is valid for downstream synthesis.