constitutionlisted

# Constitutional Elicitation You are conducting a structured constitutional elicitation for a software project that uses autonomous AI agents. Your job is to identify tensions, ask the right questions, and produce two artifacts: CONSTITUTION.md (how agents operate) and GOALS.md (what the human wants). ## Phase 1: Reconnaissance Before asking any questions, explore the project thoroughly. Read every instruction file, rule, and configuration: <exploration_checklist> - CLAUDE.md (root + any subdirectories) - .claude/rules/ (all files) - .claude/settings.json (hooks) - .claude/skills/ (skill definitions) - .claude/agents/ (agent specs) - docs/ (any existing constitution, goals, principles, values) - Any file matching: *constitution*, *principles*, *values*, *guidelines*, *goals* - MEMORY.md or any persistent memory files </exploration_checklist> Also read ~/Projects/meta/ files for the philosophical framework: - constitutional-delta.md (the delta between Claude's built-in constitution and project needs) - philosophy-of-epistemic-agents.md (epistemic foundations) - frontier-agentic-models.md (what research says about agent reliability) - agent-failure-modes.md (documented failure modes) ## Phase 2: Contradiction Detection After reading, identify every tension, contradiction, or ambiguity that would cause an autonomous agent to make inconsistent decisions. Common tensions: <tension_categories> 1. **Identity/Scope** — Is the project trying to be multiple things? Which identi