Spec-crew orchestration-governance

Governance skill for orchestrating CrewAI specialists with strict ownership, skill boundaries, and execution permissions.

install

source · Clone the upstream repo

git clone https://github.com/victorgrein/spec-crew

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/victorgrein/spec-crew "$T" && mkdir -p ~/.claude/skills && cp -r "$T/templates/shared/skills/orchestration-governance" ~/.claude/skills/victorgrein-spec-crew-orchestration-governance && rm -rf "$T"

manifest: templates/shared/skills/orchestration-governance/SKILL.md

source content

Orchestration Governance

Purpose

Route user requests to the correct specialist, enforce skill boundaries, and validate outputs before returning results.

Specialist Contracts

Specialist	Primary purpose	Allowed skills	Permission profile
builder	Build crews, agents, tasks, tools, memory	`core-build` , `tools-expert`	normal write/edit
flow	Flows, state-management, routing, orchestration, decorators	`flows`	normal write/edit
auditor	Investigation, auditing, validation	`core-build` , `flows` , `tools-expert`	read-only
docs	Documentation authoring and maintenance	`core-build` , `flows`	write/edit only `*.md` ; bash read-only

Routing Rules

Builder keywords: create, build, scaffold, setup, agent, task, crew, tool, memory
Flow keywords: flow, state, router, routing, orchestration, decorator, @start, @listen, @router
Auditor keywords: inspect, audit, investigate, trace, root cause, validate, performance analysis
Docs keywords: docs, readme, guide, documentation, architecture notes, standards

Canonical Command Ownership

```
/crew init
```
-> builder
```
/crew inspect
```
-> auditor
```
/crew fix
```
-> auditor first, then builder/flow for implementation when needed
```
/crew evolve
```
-> flow
```
/crew docs
```
-> docs

Delegation Pattern

Write plain-language delegation with:

Goal
Context and file paths
Specialist skill constraints
Deliverables
Validation criteria

Auditor Rule

Auditor executes read-only analysis and returns:

findings
risk assessment
recommendations
validation steps

Auditor does not patch files.

Validation Checklist

Chosen specialist matches intent and ownership
Delegation includes exact paths and constraints
Specialist used only allowed skills
Permission limits are respected
Output contract is complete and actionable

Supporting Files

Subagent directory:
```
references/subagent-directory.md
```
Delegation patterns:
```
references/delegation-patterns.md
```
Question templates:
```
references/questioning-guide.md
```
Delegation template:
```
assets/delegation-template.md
```