Evolution strategy supervisor
Skill panjose/Co-Scientist/skills/evolution-strategy-supervisor
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 evolution-strategy-supervisorAssembled 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
Choose exactly one concrete evolution strategy for the active evolution round.
SKILL.md
2.5 KB, as published. Nobody here has run it
evolution-strategy-supervisor
Goal:
- Choose exactly one concrete evolution strategy for the active evolution round.
Inputs:
research_plan/RESEARCH_PLAN.jsonstate/STRATEGY_PLAN.json- selected parent
hypotheses/<id>/HYPOTHESIS.jsonartifacts with completed review bundles
Outputs:
- one chosen evolution strategy for the round
- selection rationale in the skill trace
Context Loading:
- Read
research_plan/RESEARCH_PLAN.json. - Read
state/STRATEGY_PLAN.json. - Confirm that
next_actioniscontinue_evolution. - Read every parent hypothesis listed in
signals.selected_parent_ids. - Read the latest review findings for those parent hypotheses.
Execution Prompt Contract:
- System Intent:
- You are the round-level supervisor that chooses one evolution strategy from the currently allowed bundle.
- Required Reasoning Focus:
- Respect
signals.selection_strategy. - For
single_island, prefer strategies that refine one hypothesis:grounding_evolutioncoherence_evolutionfeasibility_evolutionsimplification_evolution
- For
multi_island, prefer strategies that combine or diverge across parents:inspiration_evolutioncombination_evolutionout_of_box_evolution
- Use the parent review bundle to choose the most corrective or most leverageable move for this round.
- Respect
- Do Not Do:
- Do not return multiple final strategies.
- Do not choose a strategy outside
selected_evolution_strategies. - Do not ignore the active parent set.
- Do not append, rewrite, or enrich
state/STRATEGY_DECISIONS.jsonl; this supervisor only chooses the concrete strategy for the current round. - Do not write child hypothesis IDs, tournament IDs, proximity statuses, convergence counts, or top-k entry results into the strategy decision log.
Execution Steps:
- Read the required artifacts.
- Confirm the parent set and island-selection mode for the current round.
- Inspect the review weaknesses or synthesis opportunities in the parent set.
- Choose exactly one evolution strategy from
selected_evolution_strategies. - Record the rationale in the trace and hand the chosen strategy to the evolution loop without mutating
state/STRATEGY_DECISIONS.jsonl.
Completion Rule:
- This skill is complete only when one concrete evolution strategy has been chosen for the active round and the choice is justified against the parent review context.