devils-council-opencode

Persona-driven adversarial review for plans, code, and design artifacts — as an OpenCode plugin.

Catch weak plans, overengineered designs, and business misalignment before execution — by surfacing the pushback a senior engineering org would give.

Install

Add to your project's opencode.json:

{
  "plugin": ["devils-council-opencode"]
}

OpenCode auto-installs npm plugins at startup.

Agents

| Agent | Role | |-------|------| | @staff-engineer | Pragmatist — asks what we can delete or not build | | @sre | Operational reality — asks what breaks at 3am | | @product-manager | Business alignment — asks who filed the ticket | | @devils-advocate | Premise attack — names the unquestioned assumption | | @council-chair | Synthesizes contradictions across all persona scorecards | | @council-review | Full council in one invocation (4 core + bench + Chair) | | @security-reviewer | Bench — triggers on auth/crypto/secrets code | | @finops-auditor | Bench — triggers on cloud resources/SDK imports | | @air-gap-reviewer | Bench — triggers on external dependencies/egress | | @performance-reviewer | Bench — triggers on hot-path/N+1 patterns |

Usage

Invoke any persona directly:

@staff-engineer review this plan: <paste artifact>

Or run the full council:

@council-review <paste artifact or file path>

How It Works

Core personas (Staff Engineer, SRE, PM, Devil's Advocate) always run. Bench personas auto-activate when structural signals are detected in the artifact (e.g., AWS imports trigger FinOps, crypto code triggers Security).

Each persona produces a structured scorecard with findings that cite verbatim evidence from the artifact. The Council Chair synthesizes contradictions by name without collapsing dissent into a scalar verdict.

Also Available

Claude Code plugin (16 personas, Codex CLI integration): see github.com/astrowicked/devils-council

License

MIT