Templates

Crewship provides two types of templates: crew templates (pre-configured teams with agents) and workflow templates (reusable mission execution patterns).

Crew Templates

Crew templates create a complete team in one step — a crew with pre-configured agents, roles, system prompts, and skill assignments. Deploy with crewship template deploy <slug> or the Settings UI.

Slugs, workspaces, and shadowing

A crew template is either built-in — shipped with the binary, seeded from embedded YAML, visible to every workspace — or workspace-owned, created inside one workspace and visible only there. Slugs are scoped accordingly:

Template kind	Slug must be unique across
Built-in	all built-ins (one shared catalog)
Workspace-owned	that workspace only

So two workspaces may both hold backend-team, and neither can see the other’s. A workspace-owned slug never collides with another tenant’s. A workspace template shadows the built-in of the same slug. When a workspace owns backend-team and a built-in backend-team exists, every by-slug operation in that workspace — template get, template deploy, the hire path, the onboarding wizard — resolves to the workspace’s version, and template list shows the slug once, as the workspace’s version. The more specific definition wins. This is what lets you customise a shipped template in place: adjust the roster or the prompts, save it under the same slug, and every existing reference to that slug picks up your version. Renaming would break those references instead. Other workspaces are unaffected — they keep seeing the built-in. To get the built-in back, delete the workspace template; nothing about the built-in was modified, it was only hidden.

Seed Data Crews

The built-in seed data (cmd/crewship/seeddata/builtin/crews.yaml + agents.yaml) provisions three demo crews with seven agents total:

Engineering
Quality
Ops

Slug: engineering | Icon: terminal | Color: #3B82F6 (blue)

Agent	Role	Agent Role	Profile	Skills
Alex	Engineering Lead	LEAD	FULL	network-probe, script-runner, file-crafter, routine-author
Sam	Backend Engineer	AGENT	CODING	script-runner, file-crafter, system-inspector
Robin	Frontend Engineer	AGENT	CODING	file-crafter, web-scraper

Slug: quality | Icon: shield-check | Color: #10B981 (green)

Agent	Role	Agent Role	Profile	Skills
Jordan	QA Lead	LEAD	FULL	script-runner, file-crafter, routine-author
Casey	Test & Review Engineer	AGENT	CODING	system-inspector, script-runner, file-crafter

Slug: ops | Icon: server | Color: #EF4444 (red)

Agent	Role	Agent Role	Profile	Skills
Morgan	SRE / Ops Lead	LEAD	FULL	network-probe, system-inspector, routine-author
Riley	Platform Engineer	AGENT	CODING	web-scraper, script-runner, network-probe

Every seed agent runs the CLAUDE_CODE adapter. Skill assignments are the per-agent map in cmd/crewship/seeddata/skills.go (SkillAssignments); the catalog of five installable skills — network-probe, file-crafter, web-scraper, script-runner, system-inspector — lives in builtin/skills.yaml. routine-author is a bundled first-party skill installed on server startup (not in that catalog) and is assigned to each crew’s LEAD.

Agent Configuration

Each seed agent has these settings:

Setting	Default Value
LLM Provider	ANTHROPIC
LLM Model	claude-haiku-4-5
Memory Enabled	true
CLI Adapter	CLAUDE_CODE
Timeout (LEAD)	3600 seconds (1 hour)
Timeout (AGENT)	1800-2400 seconds (30-40 min)

System Prompts

Each agent has a dedicated system prompt file in cmd/crewship/seeddata/prompts/{prompt_slug}.md (the prompt_slug matches the agent slug — e.g. prompts/alex.md). These prompts define the agent’s persona, expertise areas, communication style, and behavioral guidelines.

Workflow Templates

Workflow templates define reusable execution patterns for missions. They are defined in internal/orchestrator/workflow.go and available via the sidecar at GET /mission/templates.

Built-in Templates

Sequential

Tasks execute one after another in order. Each step depends on the previous one.

step-1 --> step-2 --> step-3

Parallel

All tasks run simultaneously, then results are aggregated by a lead agent.

task-a --|
task-b --|-> aggregate
task-c --|

Dev-Test Loop

Developer implements, tester reviews. On failure, loops back (max 3 iterations). This is the Ralph Loop pattern.

develop <--(fail)-- test
(max 3 iterations)

Pipeline

Sequential preparation, parallel work streams, final aggregation by lead.

prepare -> work-a --|
        -> work-b --|-> finalize

Template Step Structure

Each template step defines:

type TemplateStep struct {
    ID            string   // Unique step identifier
    Title         string   // Human-readable step title
    AgentRole     string   // Role-based assignment ("developer", "tester", "lead")
    AgentSlug     string   // Or specific agent slug
    DependsOn     []string // Step IDs this depends on
    MaxIterations int      // Max retry attempts (0 = no retry)
    LoopBackTo    string   // On failure, restart from this step ID
}

Using Templates

A lead agent lists the available templates, then replicates the pattern it wants by shaping the mission’s tasks — /mission/create has no template field, so the lead expresses the dependency graph directly via each task’s depends_on and max_iterations (mirrors TemplateStep.DependsOn / MaxIterations). Both endpoints are served by the sidecar on :9119 (internal/sidecar/mission.go).

# List available templates (GET /mission/templates → orchestrator.ListTemplates)
curl -s http://localhost:9119/mission/templates

# Create a mission shaped like the dev-test-loop template
curl -s -X POST http://localhost:9119/mission/create \
  -H "Content-Type: application/json" \
  -d '{
    "title": "Implement auth module",
    "tasks": [
      {"title": "Write auth handlers", "assigned_to": "sam", "max_iterations": 3},
      {"title": "Test auth flow", "assigned_to": "casey", "depends_on": ["Write auth handlers"], "max_iterations": 3}
    ]
  }'

assigned_to is an agent slug within the caller’s own crew; use assigned_to_id (agent ID) to delegate cross-crew. crew_id in the request body is ignored — the sidecar always uses the trusted IPC crew identity.

Credential Templates

The seed data also includes credential templates in cmd/crewship/seeddata/credentials.go:

Anthropic Credential

ResolveAnthropicCredential reads SEED_ANTHROPIC_API_KEY (falling back to a demo-placeholder-* value when unset) and auto-detects the credential type from the value prefix:

Prefix	Type	Env Var Name
`sk-ant-oat`	`AI_CLI_TOKEN`	`CLAUDE_CODE_OAUTH_TOKEN`
Other	`API_KEY`	`ANTHROPIC_API_KEY`

GitHub Credential (optional)

ResolveGitHubCredential seeds a credential only when SEED_GITHUB_TOKEN is set. It is type CLI_TOKEN (not API_KEY) with env var GH_TOKEN — mounted into the container as a 0400 file plus an env var so the in-container gh CLI authenticates with zero extra setup.

Google Credential (optional)

ResolveGoogleCredential seeds a credential only when both SEED_GOOGLE_EMAIL and SEED_GOOGLE_PASSWORD are set. It is named GOOGLE_API_CREDENTIALS, stored as a JSON object ({"email":…,"password":…}) under the SECRET type.

Get Started

Guides

Security

Configuration

Manifest Reference

Templates

Templates

Crew Templates

Slugs, workspaces, and shadowing

Seed Data Crews

Agent Configuration

System Prompts

Workflow Templates

Built-in Templates

Sequential

Parallel

Dev-Test Loop

Pipeline

Template Step Structure

Using Templates

Credential Templates

Anthropic Credential

GitHub Credential (optional)

Google Credential (optional)

​Templates

​Crew Templates

​Slugs, workspaces, and shadowing

​Seed Data Crews

​Agent Configuration

​System Prompts

​Workflow Templates

​Built-in Templates

Sequential

Parallel

Dev-Test Loop

Pipeline

​Template Step Structure

​Using Templates

​Credential Templates

​Anthropic Credential

​GitHub Credential (optional)

​Google Credential (optional)

Templates

Crew Templates

Slugs, workspaces, and shadowing

Seed Data Crews

Agent Configuration

System Prompts

Workflow Templates

Built-in Templates

Template Step Structure

Using Templates

Credential Templates

Anthropic Credential

GitHub Credential (optional)

Google Credential (optional)