Shell Scenario Panel

Use this skill when working inside:

• /Users/braydon/projects/experiments/ai-simulations/shell-scenario-panel

This skill is intentionally thin. The repo is the source of truth.

Canonical Files

Read only what you need, in this order:

1. CODEX.md or CLAUDE.md for runtime-specific behavior
2. prompts/moderator.md for the actual facilitation sequence
3. .claude/config.json for phase definitions and specialist rosters

Do not rely on older copied workflow descriptions when these files disagree.

Startup

For Codex sessions:

.codex/session-start.sh

To resume:

.codex/session-start.sh --scenario SCENARIO-YYYY-NNN

To create a new scenario explicitly:

.codex/session-start.sh --new

To open monitoring mode:

.codex/session-start.sh --monitor SCENARIO-YYYY-NNN

For Claude sessions, use the equivalent .claude/session-start.sh.

Expected Workflow

The current live workflow is:

0. Worldview elicitation 0a. Internal baseline 0b. Context enrichment
1. Focal question
2. Predetermined elements
3. Critical uncertainties
4. Scenario development
5. Early warning signals 6a. Impact analysis 6b. Strategy / response testing
6. Worldview integration

Important distinction:

• Phases 1-5 build the external frame.
• Phase 6a translates that frame into actor-relative impact.
• Phase 6b turns impact into preparedness, positioning, or response guidance.
• Phase 7 connects the work back to the user's worldview.

User-Facing Behavior

When the user invokes this skill, expected facilitator behavior is:

• bootstrap the session automatically
• check resources/ and offer to ingest materials
• elicit worldview before doing external analysis
• capture internal baseline before recommendations
• stop after scenarios and run impact translation before strategy
• use checkpoints after major synthesis steps
• never ask the user to run scripts manually

From the user perspective, they should mostly:

• describe the topic or decision
• answer Dr. Wells' questions
• validate syntheses at checkpoints
• refine the focal actor or response mode when impact work begins

Impact Layer

Treat impact_analysis.md as mandatory between scenarios and strategy_analysis.md.

The impact layer is:

• actor-relative, not scenario-relative
• cross-domain, not business-only
• suitable for household, personal finance, education, company, customer, or mixed actor systems

Default impact cast:

• Marisol Vega (ledger_keeper)
• Darnell Brooks (friction_mechanic)
• Nadia Rahman (dependency_cartographer)
• Dr. Imani Clarke (burden_cartographer)
• Ethan Rowe (optionality_conservator)
• Priya Desai (precedent_archivist)
• Luis Ortega (signal_mason)
• Jamie O'Sullivan (contrarian)

Overlay packs currently implemented:

• household_personal
• commercial_positioning

Use .claude/lib/select-impact-specialists.sh --list-overlays to inspect them or pass overlay ids to resolve the combined cast.

Research and Validation

• Use pp -r --no-interactive "query" --output json for web research when needed.
• Do not bypass transcript checks, metadata updates, or quality gates.
• Use .claude/lib/file-paths.sh and related validation helpers when prompts refer to enforced file outputs.

Exports

If exports are needed:

.claude/export-deck-brief.sh SCENARIO-YYYY-NNN

Preview exports locally with:

.claude/lib/serve-exports.sh SCENARIO-YYYY-NNN