Starters | Sandcaster

What Are Starters?

Starters are runnable project templates that give you a working Sandcaster agent configured for a specific use case. Instead of writing a system prompt from scratch, you pick a starter and get a ready-to-run project in seconds.

Each starter scaffolds:

sandcaster.json — pre-configured model, tools, and system prompt tuned for the use case
README.md — usage notes and example prompts specific to this starter
.env.example — the API keys this starter needs
Starter-specific assets — skill files, sub-agent configs, or example inputs as appropriate

Available Starters

Name	Description	Aliases
`general-assistant`	Flexible agent for mixed workflows — research, writing, analysis, coding	—
`research-brief`	Research a topic, compare options, or support a decision with sourced analysis	`competitive-analysis`
`document-analyst`	Review transcripts, reports, and PDFs — extract insights, summarize, flag issues	—
`support-triage`	Categorize, prioritize, and draft responses for support tickets or bug reports	`issue-triage`
`api-extractor`	Crawl API documentation and draft an OpenAPI specification	`docs-to-openapi`
`security-audit`	Structured security review using orchestrated sub-agents	—

Using a Starter

sandcaster init <starter>
sandcaster init <starter> <directory>

List all available starters:

sandcaster init --list

general-assistant

A flexible, general-purpose agent with no strong opinion about task shape. Good for exploratory work, mixed workflows, and tasks that don’t fit a more specialized template.

sandcaster init general-assistant
cd general-assistant
sandcaster "Draft a go-to-market strategy for a developer tool"

research-brief

Configured for structured research tasks: compare options, gather evidence, and produce a decision-ready brief. The system prompt emphasizes sourcing, balanced comparison, and a clear recommendation.

Aliases: competitive-analysis

sandcaster init research-brief
cd research-brief
sandcaster "Compare Stripe, Paddle, and Lemon Squeezy for a bootstrapped B2B SaaS"

Typical output: A markdown document with a comparison table, per-option analysis, and a final recommendation with rationale.

document-analyst

Built for reading and extracting insights from uploaded documents — meeting transcripts, customer interviews, annual reports, compliance documents, or PDFs.

sandcaster init document-analyst
cd document-analyst
sandcaster "Summarize the key decisions and open questions" -f board-meeting.txt
sandcaster "Extract all action items and assign owners" -f transcript.pdf

Typical output: Structured summary with extracted entities, themes, action items, or risk flags depending on the prompt.

support-triage

Processes support tickets, GitHub issues, or bug reports. Categorizes by type and severity, identifies duplicates, and drafts a triage response or escalation note.

Aliases: issue-triage

sandcaster init support-triage
cd support-triage
sandcaster "Triage these tickets and flag anything P0" -f tickets.json

Typical output: A triage report with severity classification, suggested assignees, and draft responses for each ticket.

api-extractor

Crawls API documentation pages and produces a draft OpenAPI specification. Useful for documenting undocumented APIs, migrating from legacy docs, or generating client SDKs.

Aliases: docs-to-openapi

sandcaster init api-extractor
cd api-extractor
sandcaster "Extract the OpenAPI spec from https://example.com/api/docs"

Typical output: A openapi.yaml file with endpoints, parameters, request/response schemas, and authentication info filled in from the docs.

security-audit

A multi-agent starter that runs a structured security review. Spawns specialized sub-agents for different audit dimensions (authentication, authorization, input validation, dependency analysis) and aggregates findings into a final report.

sandcaster init security-audit my-audit
cd my-audit
sandcaster "Audit the API layer in ./src/api for security issues"

Typical output: A structured security report with findings grouped by severity (critical, high, medium, low), each with a description, reproduction steps, and remediation suggestion.

Creating Custom Starters

The starter system is designed for community contribution. If you’ve built a useful configuration, consider contributing it back:

Fork the repository
Add your starter under packages/core/src/starters/
Include sandcaster.json, README.md, .env.example, and any supporting files
Open a pull request with a description of the use case

Custom starters follow the same conventions as built-in ones and can include skills, sub-agent definitions, structured output schemas, and example prompts.