Assay

SkillDev tools

Assay an experiment — deep analysis of results with cross-run comparison

Available today. Use it from your connected AI after setup.

No other account needed.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Assay skill

What this skill tells your AI

The instructions your AI receives, as published by rlacombe/distillate in .claude/skills/assay/SKILL.md and read by ahel’s review.

Test the purity of what the experimentalist agent produced. Deep analysis of an experiment's results.

The user provides an experiment name or project ID.

Get latest data: Call mcp__distillate__scan_project to ensure we have the most recent runs
Full history: Call mcp__distillate__get_project_details for complete run history with metrics
Identify the frontier: Find the best run for each metric, trace how the frontier evolved
Cross-run comparison: Use mcp__distillate__compare_runs to understand what changed between breakthrough runs
Session archaeology: If session histories exist, read them (Glob ~/.claude/projects/*<name>*/*.jsonl) to understand why the agent made its choices
Literature context: Call mcp__distillate__suggest_from_literature to compare with published results
Synthesize:
- Trajectory: how the metric evolved across runs (phases, plateaus, breakthroughs)
- What worked: techniques/configs that improved results
- What didn't: approaches that regressed or were abandoned
- Concrete next steps with specific parameter suggestions
Lab notebook: Offer to call mcp__distillate__get_experiment_notebook for the full record

Advanced