pre-mortem-discussion

Scaffold Claude Code pre-mortem discussion skill into the project you run it from.

Install / run

npx pre-mortem-discussion

What it writes

Scaffolds the /pre-mortem skill into the target project:

• .claude/skills/pre-mortem/SKILL.md
• .claude/skills/pre-mortem/agent-template.md
• .claude/skills/pre-mortem/domains.md
• .claude/skills/pre-mortem/.scaffold-meta.json

Update policy (non-destructive)

Per file, on re-run:

• Missing file: write it.
• Previously scaffolded and unmodified: safe overwrite/update.
• User-modified (or provenance unknown): do not overwrite; write *.incoming next to it.

If any *.incoming files are created, the CLI exits with code 2.

Usage

After scaffolding, run the skill inside Claude Code:

# Automatic expert selection (Mode A)
/pre-mortem <topic or file>

# Manual expert override (Mode B)
/pre-mortem <topic or file> --experts <e1>, <e2>, ..., <eN>

Examples:

/pre-mortem proposal.md
/pre-mortem proposal.md --experts security, database, frontend
/pre-mortem "OAuth2 with JWT for mobile"
/pre-mortem "OAuth2 with JWT for mobile" --experts security, api

Resume after context overflow:

If a session stops due to context overflow (common with 5 agents × multiple rounds), resume with:

/pre-mortem resume {name}

Progress is saved continuously to discussions/{name}/state.json. Only incomplete work is re-run.

Features

• Two modes: Automatic expert selection (Mode A) or manual override (Mode B)
• Exactly 3 rounds: Parallel discovery → Sequential debate → Parallel filter
• 15 domain experts: tech, security, database, api, frontend, mobile, ux, scalability, devops, infra, legal, cost, integration, data
• 4-step finalization: Creates approved.md, renames original, generates ADR, cleans up session
• Architect dialogue: Open-ended conversation to resolve all risks
• Files-first architecture: Crash recovery and resume capability

Non-goals

• Scaffold-only: does not run any sessions.
• Does not write or modify .claude/agents/*.