Scientist

Scientist v0.7.0

Scientific research suite with 195 sub-skills. Route work to the smallest suitable sub-skill, save all outputs as files, and leave a complete timestamped execution trace.

Core Rules

• Write report.md in the same language as the user's input.
• Keep all figure, chart, axis, legend, and annotation text in English.
• Save every artifact to files. Do not leave analysis, code, tables, or figures only in chat.
• Prefer the narrowest matching sub-skill instead of loading broad context.
• Final chat output should summarize saved files, not reproduce the full analysis.

Required Output Layout

Use this structure unless the task requires a stricter downstream format.

/workspace/group/
├── report.md
├── figures/
├── results/
├── data/
└── logs/
	└── process-log.jsonl

Minimum file expectations:

• report.md: title, execution timestamp, objective, methods, results, discussion, generated file inventory.
• figures/: saved plots only; never rely on interactive display.
• results/: numeric summaries, model outputs, metrics, or structured intermediate results.
• data/: processed datasets when data transformation occurs.
• logs/process-log.jsonl: append-only execution trace with one JSON object per event.

Required Execution Logging

For every task, write a timestamped process log to logs/process-log.jsonl. The log must capture the full workflow, including:

• the user prompt actually used for the run
• each selected sub-skill or external prompt template
• information passed into each skill or tool handoff
• information received back from each skill or tool handoff
• every file created, updated, or finalized
• any recovery action, retry, or fallback decision

Required event fields:

• timestamp
• phase
• event_type
• actor
• input_prompt
• skill_or_tool
• handoff_in
• handoff_out
• files_written
• status

Recommended event order:

1. run_started
2. prompt_received
3. skill_selected for each invoked sub-skill
4. handoff_started and handoff_completed for each transfer
5. file_written whenever a file is created or updated
6. report_finalized
7. run_completed

Example record:

{"timestamp":"2026-03-29T10:15:00Z","phase":"EXECUTE","event_type":"skill_selected","actor":"scientist","input_prompt":"optimize the assay design","skill_or_tool":"scientific-doe","handoff_in":{"objective":"maximize yield"},"handoff_out":{"design_type":"fractional-factorial"},"files_written":["results/doe_plan.json"],"status":"ok"}

Verification Loop

Every execution follows this loop:

1. PLAN: define objective, constraints, target outputs, and candidate sub-skills.
2. EXECUTE: run the selected analysis pipeline and save intermediate artifacts.
3. VERIFY: check outputs against scientific and formatting quality gates.
4. REPORT: write report.md in the user's language and embed saved figures.
5. LOG: finalize logs/process-log.jsonl and confirm the generated file inventory.

Quality Gates

• Figures are saved to figures/ and referenced from report.md.
• Numeric and structured outputs are saved under results/.
• report.md includes timestamp, methods, results, discussion, and file inventory.
• Figure text is English-only.
• logs/process-log.jsonl exists and includes prompt, sub-skill usage, handoff I/O, and files written with timestamps.
• No essential result remains chat-only.

Coverage

• Data analysis: Bayesian statistics, time-series, anomaly detection, causal inference.
• Life sciences: genomics, proteomics, ADMET, clinical NLP, structural biology.
• Chemistry and materials: cheminformatics, molecular modeling, docking, simulations.
• Research workflows: systematic review, hypothesis design, DOE, reproducible reporting.
• Visualization and AI: advanced plotting, dashboards, ML, transfer learning, ensemble methods.

Routing Notes

• Use scientific-deep-research when literature discovery and source synthesis dominate.
• Use the most domain-specific sub-skill available for computation-heavy tasks.
• Split work across sub-skills only when it improves correctness, reproducibility, or traceability.