Production Agent System Prompt Template
A fill-in-the-blanks system prompt for tool-using agents — role, goals, tool policy, autonomy limits, escalation rules, output contract and self-checks — following Anthropic's published guidance on building effective agents.
What it produces
The prompt
Unlock this prompt
~368 words, copy-paste ready. Pro members get this and every other prompt — plus a new drop each week.
How to use it
- Fill every bracket; delete sections that truly don't apply rather than leaving them vague.
- Use as the system prompt in the Claude API, the Claude Agent SDK, or a Claude Code subagent (.claude/agents/[name].md).
- Test with 10 realistic tasks including 3 that SHOULD trigger an approval or escalation, and tighten wording where it misbehaves.
More in MCP & Agents
Deep Research Agent with Verifiable Citations
A research-agent system prompt (extended from Anthropic's cookbook research agent) that plans sub-questions, searches, cross-checks sources and returns a cited report with confidence levels.
Set Up Your MCP Dev Stack in Claude Code: GitHub, Postgres, Playwright, Filesystem
A guided setup prompt that has Claude Code install, scope and smoke-test four real MCP servers — GitHub, Postgres MCP Pro, Playwright and Filesystem — with least-privilege defaults and a written runbook.
Exploratory QA Agent with Playwright MCP
Turns Claude Code + Playwright MCP into a methodical QA tester: it explores your app like a user, reproduces bugs with exact steps, and writes Playwright regression tests for what it finds.