devils-council

Delegation envelope for routing deep code scans to the OpenAI Codex CLI. Defines the request shape personas emit, the normalized response the conductor reconciles back into the persona's scorecard, and the complete error taxonomy. Phase 6 Security and Dual-Deploy Reviewer personas consume this unchanged. Shell-primary via `codex exec --json --sandbox read-only`; MCP deferred to v1.1.

Interactive scaffolder for custom devils-council personas. Collects voice-kit fields, validates schema, coaches voice distinctness. Writes to plugin data workspace.

Tone rubric and value-system-anchor discipline every persona preloads. Defines the voice kit (primary_concern, blind_spots, characteristic_objections, banned_phrases, tone_tags), the worked-example requirement, and why v1 ships no LLM-based tone validation. Consumed by every persona subagent via the `skills:` preload field. Companion doc: PERSONA-SCHEMA.md.

Canonical scorecard contract every persona targets and the Phase 3 review engine enforces. Defines finding fields (target, claim, evidence, ask, severity, category), the 4-tier severity enum, the verbatim-evidence grounding rule, and the structural drop rule for banned vague verbs. Consumed by every persona via the subagent `skills:` preload field.