cubetribe/claudecode_godmode-on/meta-decisions

User Prompt Received
    ↓
Orchestrator applies the meta-decision rules natively (Core Rule 3), evaluating:
    ↓
┌─ Security keywords? → securityOverride
├─ Breaking change? → breakingChangeEscalation
├─ Performance? → performanceCriticalPath
├─ Emergency? → emergencyHotfix
└─ Docs only? → documentationOnly
    ↓
None matched → Standard workflow selection

## ADR-[NUMBER]: [Title]
 
**Date:** YYYY-MM-DD
**Status:** Accepted / Superseded / Deprecated
**Context:** [Why this decision was needed]
**Decision:** [What was decided]
**Consequences:** [Impact of the decision]

START: an agent, gate, or the orchestrator hits friction
│
├─ Is it a retriable tool error or a missing input the agent can find itself?
│  └─ YES ─▶ TIER 1 (agent self-resolution)
│            - max 2 attempts, same agent, same scope
│            - attempt 2 still fails ─▶ fall through to TIER 2
│
├─ Is it a write-scope conflict, a gate failure with actionable findings,
│  or a mid-sprint routing re-classification (new risk signal appeared)?
│  └─ YES ─▶ TIER 2 (orchestrator resolution)
│            - BLOCKED (conflict)  → orchestrator resolves ownership, re-dispatches
│            - BLOCKED (quality)   → merge findings, rework loop back to @builder
│            - new risk signal     → re-route Smart Routing → Full-Gates, log why
│
└─ Is it one of the following?
   │
   ├─ MANDATORY (Tier 3 human escalation — no orchestrator override):
   │  - security-relevant finding (any agent, any tier)
   │  - scope change vs. the approved plan/sprint file
   │  - verdict conflict NOT resolvable by written criteria
   │    (see Finding-Conflict Adjudication in `docs/orchestrator/QUALITY-GATES.md`)
   │  - any destructive/irreversible action (force-push, history rewrite,
   │    data deletion, prod deploy)
   │  - release/publish steps (tag, GitHub Release, npm publish, merge to main)
   │  - architecture selection between two or more workable alternatives
   │  - design-taste decisions no written criterion can resolve (API
   │    ergonomics, public-surface naming, UX judgment)
   │  - suspicion that the request itself is malformed (requirements
   │    silently conflict, a "bug fix" is actually a design flaw)
   │  - any decision where the Finding-Conflict Adjudication procedure
   │    hit Step 3
   │
   └─ OPTIONAL (orchestrator judgment — permitted to proceed without asking,
      but MUST be logged in the sprint's Routing Log):
      - cost overruns (token/time budget exceeded)
      - repeated rework loops (>2 cycles through Tier 2 on the same item)
      - ambiguity that slows progress but does not block a gate or a write

{
  "type": "feature|bug|enhancement",
  "complexity": "low|medium|high",
  "areas": ["api", "ui", "backend"],
  "meta_rules_triggered": ["securityOverride"],
  "workflow_adaptation": "Added @security check"
}

User: "Hotfix: production login broken"
    ↓
@builder (immediate fix)
    ↓
Deterministic hook (typecheck/lint/tests/build)
    ↓
@tester and/or @security skipped IF the sprint's risk profile allows it —
logged in the Routing Log with the reason, not silently dropped
    ↓
@scribe adds the `[Unreleased]` CHANGELOG entry — NOT skippable, even for a
one-line hotfix (`docs/orchestrator/VERSIONING.md`: "No exceptions")
    ↓
DONE