experiment-analyst

name	experiment-analyst
description	Use when you want Cursor Agent to turn CSV results under assets/results/ into structured observation cards in assets/notes/agent_artifacts/observations/observation_output.json.

Experiment analyst

This skill is for extracting observation cards, not for writing the Experiments section.

Use it when:

• the goal is to turn result files into candidate observations
• you will write or edit Experiments prose yourself
• the model should behave like an analyst, not an author

Do not use it to:

• draft the full Experiments section without human review
• approve observations automatically
• write explanation outside the artifact file

Output contract

Write exactly one JSON artifact to:

assets/notes/agent_artifacts/observations/observation_output.json

Read the schema in references/observation-output-schema.json before producing output.

Allowed top-level shapes:

{"observations": []}

or a bare JSON array. If a field is unknown, leave it empty instead of inventing it.

Workflow

1. Read assets/notes/project.md if it exists.
2. Read result files under assets/results/.
3. Read references/observation-output-schema.json.
4. Produce candidate observation cards only.
5. Save them to assets/notes/agent_artifacts/observations/observation_output.json.
6. Stop.

Constraints

• Keep status as candidate.
• Use short, explicit claim_candidate text.
• Keep numbers ordered as baseline, ours, delta when possible.
• Use is_main_text_worthy conservatively.
• Keep warning empty when there is no boundary note.
• Do not include Markdown, comments, or prose in the artifact.

Prompts

For Codex, reuse references/codex-prompt.md.

For Claude, reuse references/claude-prompt.md.