Team Collaboration

How a team shares a TARX project across roles — covering project setup, member onboarding, role assignment, a shared agent library, shared project keys, and lightweight governance. The example below uses a small team, but the same model scales to any size or domain.

---

The Team (example)

Person	Responsibility	TARX Role
Sam (owner)	Oversees everything, owns the account	Project Admin
Jordan (lead)	Designs workflows, manages prompts	Project Editor
Casey (operator)	Runs workflows, reviews output	Project Editor
Riley (reviewer)	Reviews and approves output	Project Reader
Stakeholder	Views final output	Project Reader (optional)

---

Step 1: Sam Creates the Project

Sam creates the agency's project:

1. Settings → Projects → New Project
2. Name: "Acme Agency — Content"
3. Slug: acme-content
4. Create

Project URL: https://app.tarx.io/sam-agency/acme-content/workflows

Sam is the Admin (project owner).

---

Step 2: Sam Adds the Project's LLM Keys

As the project admin, Sam adds the agency's LLM keys to the project. Because LLM keys are project-scoped, every member can use them immediately — nobody else has to set up keys:

Settings → Keys & Secrets → Project LLM Keys:

• Anthropic Agency Key — Claude Sonnet for complex tasks
• OpenAI Agency Key — GPT-4o-mini for routine tasks
• Gemini Agency Key — For large-context work

BYOK keys are project-scoped

LLM keys belong to the project, not to an individual. Once Sam (an admin) adds them, all members can run agents that use them — a new teammate is productive on day one without recreating keys. Provider usage bills to the agency's underlying provider accounts regardless of who triggers a run, which is exactly what an agency wants. Only admins can add or delete keys; editors and readers use them. See BYOK Keys.

---

Step 3: Sam Invites Team Members

Settings → Projects → Acme Content → Members:

Invite	Email	Role	Why
Jordan	jordan@acme.agency	Editor	Designs workflows, needs full create/edit
Casey	casey@acme.agency	Editor	Runs pipelines, needs to create test agents
Riley	riley@acme.agency	Reader	Reviews output, triggers workflows, no editing

Each person receives an invitation email. They sign up for TARX (or sign in if they have an account) and accept the invitation.

---

Step 4: Sam Builds the Shared Agent Library

Sam creates the core agents that the whole team will use:

Content Writer Agent

Name: content-writer
Model: claude-sonnet-4-6 (project Anthropic key)
Skill: Company Voice Writer (custom skill Sam creates)
Temperature: 0.7

Sam creates a Company Voice Writer custom skill encoding the agency's brand voice guidelines — once configured, any agent using this skill automatically follows the style guide.

Content Editor Agent

Name: content-editor
Model: claude-haiku-4-5 (project Anthropic key)
Temperature: 0.3

SEO Analyzer Agent

Name: seo-analyzer
Model: gpt-4o-mini (project OpenAI key)
Skill: SEO Content Writer (built-in)
Temperature: 0.2

Researcher Agent

Name: researcher
Model: claude-sonnet-4-6 (project Anthropic key)
Capabilities: web_search ✅, web_scraper ✅
Temperature: 0.5

---

Step 5: Jordan Builds the Core Workflows

Jordan (Editor role) builds the main workflow templates:

Workflow 1: Blog Production Pipeline

[Webhook: content brief] → [researcher] → [content-writer] → [content-editor] 
→ [seo-analyzer] → [HiL: Riley reviews] → [Output: final post]

Workflow 2: Social Media Batch

[Manual: blog URL] → [researcher: summarize post] → [Loop: 5 platforms] 
→ [content-writer: platform-specific post] → [Output: all posts]

Workflow 3: Client Content Calendar

[Schedule: every Monday 8am] → [researcher: trending topics in client's industry] 
→ [content-writer: 10 topic ideas] → [HiL: Casey reviews] → [Output: calendar]

---

Step 6: Casey Tests and Runs Workflows

Casey (Editor role) can:

• View all agents and their configurations
• Run any workflow (trigger manually, respond to webhooks)
• Approve/reject HiL nodes
• Create new agents for testing (using the project's shared LLM keys)
• Create test workflows

Casey cannot:

• See any LLM key's raw value (values are never shown to anyone after creation)
• Add or delete LLM keys or project credentials (admin only)
• Invite or remove team members

Casey tests each workflow by running it with sample content briefs and verifying output quality.

---

Step 7: Riley Approves Content

Riley (Reader role) has a simple interface:

1. Riley gets email notifications when the blog production pipeline pauses at the HiL node
2. Riley opens Bridge (or the Executions page)
3. Reviews the produced content
4. Clicks Approve (with an optional note) or Reject (with revision feedback)

Riley cannot create or edit agents or workflows — pure review role.

---

Step 8: Adding a Client

For a client who wants visibility into their content pipeline:

Sam invites the client's point-of-contact as a Reader:

• They can view execution history for their content
• They can approve/reject HiL nodes for their content
• They cannot see other clients' workflows or data
• They cannot modify anything

One project per client

For full isolation between clients, create a separate project per client. This gives each client their own URL namespace (/{sam-agency}/{client-name}/), their own member list, and their own execution history.

---

Role Permissions Reference

A quick reference for the agency context:

Action	Sam (Admin)	Jordan (Editor)	Casey (Editor)	Riley (Reader)	Client (Reader)
Create/edit agents	✅	✅	✅	❌	❌
Create/edit workflows	✅	✅	✅	❌	❌
Run workflows	✅	✅	✅	✅	✅
Approve HiL nodes	✅	✅	✅	✅	✅
View executions	✅	✅	✅	✅	✅
Use project LLM keys (at runtime)	✅	✅	✅	✅	✅
Add/delete project LLM keys	✅	❌	❌	❌	❌
Manage project credentials	✅	❌	❌	❌	❌
Invite members	✅	❌	❌	❌	❌
Remove members	✅	❌	❌	❌	❌
Delete agents/workflows	✅	✅	✅	❌	❌
Configure MCP servers	✅	❌	❌	❌	❌

---

Team Workflow Governance Tips

Agent Naming Conventions

Establish a naming convention before you have 50 agents:

{role}-{specialization}

Examples:

• writer-blog — Blog post writer
• writer-social-linkedin — LinkedIn-specific writer
• editor-technical — Technical content editor
• researcher-competitor — Competitor research specialist
• analyzer-seo — SEO analysis

Version Production Agents

When making significant changes to a production agent, create a copy first:

1. Copy content-writer → content-writer-v2
2. Update content-writer-v2
3. Test with Jordan and Casey
4. When satisfied, update workflows to use content-writer-v2
5. Delete content-writer (old version)

This prevents breaking running workflows during updates.

Workflow Folder Organization

Jordan organizes workflows into folders:

• Production — Live, client-facing workflows
• Development — Work in progress
• Archive — Deprecated workflows kept for reference

Only promote to Production after testing in Development.

Bridge for Team Communication

Sam uses Bridge to quickly answer team questions:

• "What system prompt does the content-writer use?" → Bridge shows the current prompt
• "Which workflows use the researcher agent?" → Bridge lists them
• "How many executions did we run this month?" → Bridge queries execution stats

Bridge is the team's AI administrator — reducing the need for Sam to answer routine configuration questions.