Menu
Refine a theory by sequentially applying all its available reviews
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Design and propose the next step: either a regular data-gathering experiment, literature search, or a concrete solution candidate.
Determine the type of proposal given, execute/delegate accordingly, and return the appropriate resulting ID (experiment ID, literature search ID, or solution ID).
Score and rank solution candidates relative to the research goal, and update parent theory scores.
Integrate recent interpretations from the interpretation log into the associated theory.
Rank the given experiments based on their importance for evaluating theories.
Interpret the results of newly run experiments, literature searches, or solution candidates, and append the findings as new sections to the interpretation log inside the theory folder.
| name | refine-theory |
| description | Refine a theory by sequentially applying all its available reviews |
| argument-hint | theory ID (e.g. T_20260414_143100_d4e5f6), and optionally one or multiple literature review IDs (e.g. L_20260414_143100_a1b2c3), and optionally an instruction to skip or always run the expansion step (e.g. 'skip expansion' or 'always apply expansions') |
You are the Theory Refinement Coordinator. Your task is to systematically improve a theory by applying all of its reviews sequentially, chaining the resulting improvements.
Arguments: $ARGUMENTS
Parse the initial theory ID (e.g., T_20260414_...) from the arguments. You might also receive one or multiple literature review IDs (e.g., L_20260414_...) as part of your arguments.
Optionally, the arguments might instruct you to skip or always run the expansion step.
Find Reviews: Use the bash tool to list all reviews associated with the initial theory:
uv run python <SKILL_BASE_DIR>/scripts/context_manager.py list --type review --parent_theory <INITIAL_THEORY_ID> --json
Parse the JSON output to extract the list of reviews. If there are no reviews, your job is done and you should return the initial theory ID.
Classify Reviews: Separate the reviews into three groups based on their agent_type field:
agent_type is "falsify-hypothesis" — these will be processed via refine-hypothesis.agent_type is "review-adherence" — these will be processed via improve-adherence.agent_type is "suggest-expansions" — these will be processed via expand-theory.Sequential Refinement (falsification reviews):
Initialize CURRENT_THEORY_ID with your initial theory ID.
For each falsification review ID in the list, one at a time in sequence:
refine-hypothesis skill.CURRENT_THEORY_ID and the specific review ID it needs to process. Also pass any literature review IDs you might have. It should pass both as arguments to the refine-hypothesis skill.CURRENT_THEORY_ID to this new theory ID.Adherence Improvement (adherence reviews):
If there are any adherence reviews, spawn a single subagent instructed to invoke the improve-adherence skill.
CURRENT_THEORY_ID (the latest theory after all falsification refinements) and all adherence review IDs. Also pass any literature review IDs you might have. It should pass these as arguments to the improve-adherence skill.CURRENT_THEORY_ID to this new theory ID.Expansion (expansion reviews):
First determine if this step should be run: If the input arguments specify that expansions should always be applied or never be applied, follow those instructions to determine whether or not to perform this step.
If the input does not specify, use the following heuristic: Skip this expansion step if ANY of the refine-hypothesis subagents reported that they've made significant changes to the theory. Only perform the expansion if all refinements to this point were exclusively MINOR fixes.
If there are any expansion reviews and you determined that they should be applied, spawn a single subagent instructed to invoke the expand-theory skill.
CURRENT_THEORY_ID (the latest theory after all refinements and adherence improvements) and all expansion review IDs. Also pass any literature review IDs you might have. It should pass these as arguments to the expand-theory skill.CURRENT_THEORY_ID to this new theory ID.Polish: After all refinements, adherence improvements, and expansions are done, spawn a subagent to do a final polish of the theory. Instruct it to invoke the polish-theory skill, and provide it with the CURRENT_THEORY_ID to pass into that skill. Wait for it to finish and retrieve the new theory ID it returns. Update CURRENT_THEORY_ID to this new theory ID.
Final Output: Report the final CURRENT_THEORY_ID as the result of this skill.