From edgedelta-teammates
Guides creation and configuration of EdgeDelta AI teammates. Use when users need to build teammates, write system prompts, assign connectors, or scope permissions.
How this skill is triggered — by the user, by Claude, or both
Slash command
/edgedelta-teammates:edgedelta-teammatesThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Creates, configures, and manages EdgeDelta AI teammates. Provides guidance on system prompt engineering, connector permissions, channel scoping, and teammate lifecycle management.
Creates, configures, and manages EdgeDelta AI teammates. Provides guidance on system prompt engineering, connector permissions, channel scoping, and teammate lifecycle management.
Activate this skill when the user:
For fast lookups without loading full skill context:
# Teammate configuration fields
grep -n "Configuration Fields" SKILL.md
# System prompt patterns
grep -n "Prompt Structure" SKILL.md
# Permission model
grep -n "Permission" SKILL.md
# Specialized teammates
grep -n "Specialized" SKILL.md
# Detailed prompt engineering guide
cat assets/references/prompt-engineering.md
# Teammate data model reference
cat assets/references/teammate-model.md
When: User wants to build a new teammate from scratch
Steps:
Design the persona - Ask three questions before building:
Gather required inputs:
^[a-zA-Z0-9\s_-]+$)Build the system prompt - Construct three prompt layers:
assets/references/prompt-engineering.md for detailed patternsAssign connectors - Select minimum required connectors:
edgedelta-mcp and edgedelta-documentation auto-attached by defaultAssign to channels - Scope availability:
#incident-response, #platform-ops, etc.)Save and verify - Teammate appears in the Teammates list, is available for DMs and channel interactions
When: User needs help crafting or refining teammate prompts
Steps:
Read assets/references/prompt-engineering.md for the full guide
masterPrompt structure - Build in this order:
Identity: Who is this teammate and what is their expertise?
Responsibilities: What specific tasks does it handle?
Constraints: What should it NOT do?
Communication style: Tone, format, verbosity
Tool usage patterns: When/how to use available connectors
Escalation rules: When to ask for human approval or defer
userPrompt structure - Wrap the user's question with context:
Available template variables:
- {{question}}: The user's actual question
- {{memory_context}}: Retrieved conversation memories
- {{shared.current_agent.full}}: Full agent object
- {{shared.current_agent.reference}}: Agent id, name, description, type
Validate templates - All prompts are compiled as Handlebars templates at save time. Invalid syntax is rejected.
Test and iterate - Chat with the teammate, refine based on:
When: User wants to restrict what a teammate can see and do
Steps:
Connector-level scoping: Only assign connectors the teammate needs
Tool-level scoping: Within each connector, enable/disable individual tools
unsupervised: Executes automaticallysupervised: Requires human approval before executionChannel scoping: Limit where the teammate operates
Model constraints:
When: User wants to modify a built-in specialized teammate
Steps:
Built-in specialists (maintained by EdgeDelta):
| Teammate | Domain | Recommended Connectors |
|---|---|---|
| OnCall AI | Orchestration & routing | None (orchestrates others) |
| Cloud Engineer | Infrastructure & cost | AWS, Azure, GCP |
| Code Analyzer | Code quality | GitHub |
| DevOps Engineer | Reliability & CI/CD | Kubernetes, GitHub, Jenkins, CircleCI |
| Issue Coordinator | Workflow sync | GitHub, Jira, Linear |
| Security Engineer | Security posture | GitHub, AWS, Jira, CrowdStrike FDR |
| SRE | Incident response | EdgeDelta MCP, PagerDuty, Kubernetes |
When: User needs to update, clone, deactivate, or delete a teammate
Steps:
Update: Edit configuration, prompts, or connectors
Clone: Duplicate a teammate for staging or regional variants
Deactivate: Toggle Online/Offline without deletion
Delete: Permanently remove
When: User wants a teammate to perform automated recurring checks
Steps:
| Field | Rules | Example |
|---|---|---|
name | 1-48 chars, ^[a-zA-Z0-9\s_-]+$ | Kubernetes SRE |
masterPrompt | Valid Handlebars template | You are a Kubernetes specialist... |
userPrompt | Valid Handlebars template | Context: {{question}} |
status | active, inactive, error | active |
| Field | Rules | Default |
|---|---|---|
description | Free text | None |
avatar | Data URL or hosted URL | None |
model | Org-allowed model name | Org default |
modelTemperature | 0.1-1.0 | 0.1 |
toolingPrompt | Valid Handlebars template | Auto-generated |
toolConfigurations | Map of connector tools | edgedelta-mcp + edgedelta-documentation |
capabilities | Array of capability strings | None |
expertise | Array of expertise strings | None |
priority | Numeric priority | None |
| Type | Purpose | Created By |
|---|---|---|
user-defined | Custom teammates you create | Users via UI/API |
default | Built-in specialists (SRE, Security, etc.) | EdgeDelta (auto-backfilled) |
internal | System teammates | EdgeDelta platform only |
super-agent | OnCall AI orchestrator | EdgeDelta platform only |
Only user-defined and default are visible to normal users. internal and super-agent are restricted.
Allow (unsupervised): Execute without approvalAsk Permission (supervised): Human must approve before executionUser: "Create a teammate that monitors our Kubernetes cluster health" Agent: Uses Workflow 1, designs K8s-focused persona, assigns EdgeDelta MCP + Kubernetes connectors, builds system prompt with incident detection focus
User: "My teammate gives generic answers about security" Agent: Uses Workflow 2, refines masterPrompt with specific security tasks, decision criteria, tool usage patterns, and escalation rules
User: "How do I restrict what my DevOps teammate can do?" Agent: Uses Workflow 3, explains connector/tool/channel scoping, helps set supervised execution for write operations
User: "I want the SRE to also check PagerDuty incidents" Agent: Uses Workflow 4, adds PagerDuty connector to SRE specialist, configures tool permissions
User: "Set up a daily health check report" Agent: Uses Workflow 6, creates periodic task with daily schedule, assigns to appropriate channel
When: Teammate needs pipeline data access (streaming connectors feed EdgeDelta MCP)
When: Teammate needs to reference or create dashboards
When: Building a teammate that works with pipeline configuration
edgedelta-pipelines skilledgedelta-dashboards skilledgedelta-reference skilledgedelta-ottl skillYou are an expert in EdgeDelta teammate configuration. Guide users through persona design, system prompt engineering, connector scoping, and permission management. Always recommend starting narrow and granting least-privilege access. For prompt refinement, read assets/references/prompt-engineering.md. For data model details, read assets/references/teammate-model.md.
npx claudepluginhub edgedelta/claude-code-plugins --plugin edgedelta-teammatesReturns the universal governance specification for swarm workflow commands: hard rules, briefing templates, launch mechanics, and pulse setup. Useful when a user-authored shortcut command needs the full spec.
Creates and manages dynamic teams of domain-specific agents. Analyzes project to propose 5-20 agents with tracking framework. Modes: create, update, status, cleanup.
Structured messaging protocols for agent team communication: message type selection, plan approval, shutdown procedures, and anti-patterns to avoid.