Menu
Convert an exported chat YAML into a bot scenario file.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Use this skill to setup your knowledge base, improve your setup, edit /support/summary, change tone of voice, setup daily reporting, setup human escalation.
Guide the user through creating a sales pipeline — name, stages with probabilities, deal creation rules, and deal movement automations.
Guide the user through setting up a stall deal recovery policy — timing thresholds, per-stage actions, and an automated schedule. Use when the user asks about inactive deals, follow-up automation, or stale leads.
Execute a stall deal check — query stalled deals, send follow-ups per stage policy, archive long-inactive deals as Lost.
Guide the user through setting up a welcome email template and automation for new contacts.
Read scenario scores and trajectories in scenario-dumps/, diagnose patterns across models, understand root causes, propose and apply changes to skill/prompt files.
| name | make-scenario |
| description | Convert an exported chat YAML into a bot scenario file. |
Convert a real chat YAML into a scenario for bot testing.
Read about scenario naming convention in AGENTS.md.
User usually provides a chat YAML file path, e.g. flexus_simple_bots/karen/chat_xxx123.yaml.
Read it fully before doing anything. Also read the bot's _install.py to understand expert names —
you need them to pick the correct scenario file name prefix. Copy it using cp to the right place,
and then do some edits.
Scenarios are idealized examples of behavior. A judge will compare ideal and real behavior.
Judge needs to see unnecessary steps against the background of minimal number of steps. So your goal is to delete unnecessary steps, or information that later turns out to be false.
Maybe you can improve wording of assistant having the benefit of hindsight (you see future steps and hints, assistant didn't at the time)
Remove role=hint messages if any. They are part of scenario generating process to steer the model.
Remove skill text from flexus_fetch_skill tool calls. Replace the full content with <explanation> (unquoted in YAML).
Just makes the scenarios smaller with no loss because there's skill text right there in the same repo.
Remove dead-end explorations and failed calls — delete both the call and its tool response. Don't leave orphan references to removed things in assistant text. Remember, we are making idealized trajectory here.
Remove extra tasks on the kanban board that does not participate in the decision process later.
Most of judge instructions are already in the judge system prompt. Only very specific things about a particular scenario are needed, and most of the time you don't know what they are until you try running a judge on real trajectories.
It's something like "relax don't penalize this much", "omg humans really don't like that". Comes from experimentation.
So write "No additional instructions" as a placeholder. The bot author might edit it later.
