Steer

SkillAI & models

Steer a experimentalist agent — review progress and redirect a running experiment

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 Steer skill

What this skill tells your AI

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

Review what a experimentalist agent is doing and redirect it. Like an alchemist adjusting the flame under the athanor.

The user provides an experiment name and optionally a steering direction.

Check status: Call mcp__distillate__manage_session with action="status" to see if the experimentalist agent is running
Review progress: Call mcp__distillate__get_project_details to see recent runs, metrics, and trajectory
Read the experimentalist agent's mind: If session history exists (Glob ~/.claude/projects/*<name>*/*.jsonl), read the most recent session to understand:
- What the agent is currently trying
- What it learned from recent runs
- Whether it's stuck in a loop or making progress
Gather intelligence (if user hasn't specified direction):
- Check if goals are being approached or stalled
- Call mcp__distillate__suggest_from_literature for paper-inspired ideas
- Look at dead ends from prior sessions to avoid repeating them
Write steering instructions: Call mcp__distillate__steer_experiment with specific, actionable guidance:
- Reference what the agent has already tried (to avoid repetition)
- Suggest concrete next steps with parameter values
- Set priority: what to try first vs. fallback options
Confirm: Report the steering instructions and whether the experimentalist agent picked them up (the agent reads .distillate/steering.md via the post_bash hook)

Advanced