# Crucible

> Five-phase adversarial review protocol. PRISM finds issues. RED VIPER proves them. ASSAYER verifies alignment. Evidence decides.

- Skill: `cogpros/crucible` (Agent Skill)
- Install (CLI): `npx skillmds@latest add cogpros/crucible`
- Raw SKILL.md: https://api.skillmd.com/api/skills/cogpros/crucible/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: cogpros (https://skillmd.com/u/cogpros)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/cogpros/crucible

---


# CRUCIBLE -- Five-Phase Adversarial Review

CRUCIBLE extends PRISM with empirical exploitation testing and alignment verification. Opinions identify risks. Evidence decides them.

## When to Use

- Shipping anything where quality matters
- Adversarial testing before release
- Security-critical changes
- Public-facing content or code

## Phases

1. **PRISM (Phases 1-2)** -- Parallel specialist review (delegates to prism.sh)
2. **RED VIPER (Phase 3)** -- Adversarial exploitation testing. Proves findings, not just opines.
3. **ASSAYER (Phase 4)** -- Alignment verification against stated goals
4. **VERIFY (Phase 5)** -- Re-run phases 3+4 if needed. Confirm fixes hold.
5. **Synthesis** -- Consolidate all phases into final verdict

## How to Execute

### Script-based (portable, any agent):
```bash
~/.openclaw/scripts/crucible.sh --artifact <path> [--mode standard|dissertation|wcb-legal]
```

### Flags:
- `--skip-viper` -- Skip RED VIPER phase
- `--skip-assayer` -- Skip ASSAYER phase
- `--viper-only` -- Run only RED VIPER
- `--assayer-only` -- Run only ASSAYER
- `--gate` -- Hard gate mode (blocks on failure)
- `--dry-run` -- Show what would run without executing

## Dependencies

- Script engine: `~/.openclaw/scripts/crucible.sh`
- Shared lib: `~/.openclaw/scripts/lib/review-common.sh`
- PRISM: `~/.openclaw/scripts/prism.sh`
- RED VIPER: `~/.openclaw/scripts/red-viper.sh`
- ASSAYER: `~/.openclaw/scripts/assayer.sh`
- Sentry: `~/.openclaw/scripts/sentry-check.sh`
- Prompt templates: `$WORKSPACE/.claude/skills/prism/prompts/`

## Error Recovery

- If PRISM phase fails, crucible exits (phases 3-5 depend on PRISM findings)
- If RED VIPER fails, synthesis proceeds with PRISM + ASSAYER findings
- If ASSAYER fails, synthesis proceeds with PRISM + RED VIPER findings
- Budget limits per phase: RED VIPER $1.00, ASSAYER $0.80 (configurable)