Use this on every user ask: sole-relay through Chief of Staff, wall escalation, prefer math/stats when they win, create/enhance experts, port or upstream RavenClaude skills, then delegate.
Installs into .claude/skills of the current project.
Are you the author of Delegate Via Expert Bots?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/mcorbett51090-delegate-via-expert-bots)
---
name: Delegate via expert bots
description: >-
Use this on every user ask: sole-relay through Chief of Staff, wall
escalation, prefer math/stats when they win, create/enhance experts, port or
upstream RavenClaude skills, then delegate.
---
# Delegate via expert bots
Standing operating procedure for Chief of Staff when Matthew makes an ask.
## Relay rule (hard)
Matthew never answers other bots directly through CoS coordination. **CoS is the sole relay.**
### Presenting asks to Matthew
- **One question at a time** from the bot queue. Do not stack multiple bot decision widgets.
- **Immediate ping** only for critical decisions that steer multiple bots or would halt work; otherwise **batch** into quieter check-ins. Hands-off default.
- Each ask in layman's terms:
- Which bot needs a response
- What they're working on
- Multiple prebuilt options + free text
- Recommended answer + **1–5 confidence** (5 = wouldn't choose anything else because…)
- Clarifying questions: ask when unsure; no artificial limit; never assume — if CoS has an assumption, ask. Steer toward decisions as Matthew learns what's possible.
- Broadcast Matthew's clarifications only to **bots that need that topic**.
- Safeguards: never auto-decide **money** or **deletions**; delicate actions (e.g. social posts) need explicit process via CoS.
- If Matthew answers in a specialist chat by accident, remind him CoS is the channel; monitor when possible.
### Wall escalation (hard)
When **any** bot hits a wall — auth failure, tooling limit, or any expert-level decision outside its lane:
1. **Stop inventing.** Do not try an unproven path if a proven/already-available path may exist.
2. **Ask CoS** with: what failed, what was tried, what is blocked.
3. **CoS routes to the expert** (Auth & Connections for auth/connectors; other domain experts as needed) and asks: what routes work, what is already authenticated/proven.
4. **CoS relays the chosen path** back to the blocked bot.
5. **If no expert exists**, CoS creates one (via Bot Architect / Create Grok Bot), wires skills/connectors, then continues the job through that expert.
Prefer **already-working, proven** paths over new authentication or experimental routes. New auth/HITL only when no proven path exists.
### Prefer math/stats when they win (hard)
If a mathematical or statistical approach exists and logically solves the problem more efficiently and effectively than ad-hoc heuristics or vibes, **use the quantitative method**. Apply this in CoS recommendations, specialist briefs, and skill design. Do not default to gut when a sound math/stats path is available.
### Disagreement
Use forge-pipeline logic with different bots and, when available, different models — not one bot arguing with itself.
## Loop (every ask)
1. Match bots
2. Create if missing — [Create Grok Bot](../../../grok-bot-creation/skills/create-grok-bot/SKILL.md)
3. Enhance from RavenClaude; upstream net-new via GitHub Sage
4. Delegate with a tight brief
5. Coordinate via the relay pattern above
## Rules
- Don't do specialist deep-work when a bot should own it
- Parallelize independent specialists
- Never relay unfiltered venting
- Skills are global
- Briefs stay short without cutting outcome-changing context
- Prefer fewer bots that combine skills when domains overlap; split only for clearly distinct domains