القائمة
Intelligent routing between Solara and Codex — decides when to delegate backend tasks to Codex vs handle internally.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Mandatory compliance gate before any outbound (email, SMS, blast). Verifies consent, opt-out status, CASL footer, send window, and daily cap.
Queue a cold outreach campaign to a segmented lead list via the cold_outreach_campaigns endpoint, with CASL compliance enforced before any send.
Generates the structured daily call sheet for Ezra — 6 prioritized categories, specific opening lines for each call, and a clear action order.
Surfaces overdue follow-ups, prioritizes the call list, and recommends specific outreach actions for each stuck deal or contact.
Domain literacy for MCA and alternative business funding. Reference glossary and clarifying-question guide for when operators use ambiguous terms.
Uses historical lender feedback data to recommend the best lenders for a specific deal profile and surfaces typical terms and decline patterns.
| name | codex-delegation |
| description | Intelligent routing between Solara and Codex — decides when to delegate backend tasks to Codex vs handle internally. |
| tags | ["skill"] |
| triggers | ["codex delegation","use codex delegation","run codex delegation"] |
| tier | stable |
Purpose: Solara and Codex are complementary. This skill determines WHEN and HOW to delegate work to Codex for maximum leverage.
Solara (Claude) Codex (GPT-5.4)
├── Funding workflow orchestration ├── Backend API implementation
├── Lender intelligence synthesis ├── Deep debugging with stack traces
├── Operator handoff + comms ├── Adversarial code review
├── Memory / state management ├── Parallel backend tasks
├── CASL compliance judgment ├── Root-cause analysis
└── Call sheet + daily ops └── Write-capable rescue tasks
| Task Type | Why Codex |
|---|---|
| Pre-ship code review of API changes | Second pair of eyes catches blind spots |
| Backend bug with stack trace in logs | Codex excels at systematic root-cause |
| Implementing a new API endpoint | Runs in background while Solara works on ops |
| Test suite debugging | Codex is strong at test diagnosis |
| Database query optimization | Codex handles SQL + performance analysis well |
| Task Type | Why Solara |
|---|---|
| Funding workflow decisions | Requires domain context + operator relationship |
| CASL compliance judgment calls | Regulatory + business context needed |
| Lender relationship intelligence | Solara has the historical context |
| Operator handoff messages | Must match Ezra's communication style |
| Memory/state/context files | Solara's domain — Codex has no access |
| Simple fixes (< 3 files) | Delegation overhead > task effort |
export CLAUDE_PLUGIN_ROOT="/c/Users/User/.claude/codex-plugin"
node "$CLAUDE_PLUGIN_ROOT/scripts/codex-companion.mjs" task --write \
"Context: SunBiz-Agent — Next.js 14 App Router, Supabase, TypeScript.
Stack: scripts/integrations/send_gateway.py is the single outbound send path.
Task: [specific task description]
Constraint: [any constraints]"
Always inject context: stack, file paths, constraints, what Solara has already done.
Before delegating, verify Codex CLI is operational:
codex --version 2>&1 | head -1
If command not found: npm i -g @openai/codex@latest
If version is stale (symptoms: every model alias rejected, companion loops with no progress): upgrade first.
--model spark (simpler tasks) or --model gpt-5.4-mini (faster)memory/MISTAKES.md with what Codex struggled with.Never retry the same prompt 3 times unchanged.
After any task with ≥3 commits, ≥5 files touched, or any user-facing change:
node "$CLAUDE_PLUGIN_ROOT/scripts/codex-companion.mjs" review --wait
Present both reviews verbatim to Ezra — Solara's self-review first, then Codex's audit under a clear header. Do not paraphrase or soften Codex findings.
# Standard review
node "$CLAUDE_PLUGIN_ROOT/scripts/codex-companion.mjs" review --wait
# Adversarial design challenge
node "$CLAUDE_PLUGIN_ROOT/scripts/codex-companion.mjs" adversarial-review --wait "challenge the lender-match scoring logic"
# Delegate a task
node "$CLAUDE_PLUGIN_ROOT/scripts/codex-companion.mjs" task --write "<context + task>"
# Check status
node "$CLAUDE_PLUGIN_ROOT/scripts/codex-companion.mjs" status
# Get results
node "$CLAUDE_PLUGIN_ROOT/scripts/codex-companion.mjs" result