AgentConn
A

agency-agents

Productivity Free

About agency-agents

agency-agents (msitarzewski/agency-agents) is a curated bundle of opinionated agent persona templates packaged as markdown files for use with Claude Code, Cursor, Codex, and any harness that respects SKILL.md or per-role agent definitions. Rather than ship a runtime, the project ships *recipes* — engineer, PM, designer, marketer, sales-rep, ops, and research personas with calibrated system prompts, tool permissions, and decision guardrails. The repo crossed 92,363 total stars on May 4, 2026 (+828 in 24h), making it one of the longest-tail agent-distribution wins on GitHub: nine months of steady accumulation rather than a single viral spike. For teams that don't want to author their own personas from scratch, this is the fastest on-ramp to a multi-role agent setup.

Key Features

  • Drop-in markdown personas for engineer, PM, designer, marketer, sales-rep, ops, researcher
  • Calibrated system prompts with tool permissions per role
  • Compatible with Claude Code, Cursor, Codex, and any SKILL.md-aware harness
  • Long-tail community traction — 92K+ stars, +828 in 24h
  • Shell-based installer for one-command persona drop-in
  • Open source, MIT licensed
  • Persona definitions are auditable markdown — no opaque runtime
  • Pairs naturally with vertical-agent frameworks for human/agent-team mixed workflows

Overview

agency-agents bundles a set of opinionated agent personas — engineer, PM, designer, marketer, sales-rep, ops, researcher — as drop-in markdown files. Each persona ships with a calibrated system prompt, a tool-permission template, and decision guardrails specific to its role. Drop the files into your ~/.claude/agents/ (or the equivalent for Cursor / Codex), and the harness can route work to a named persona.

The thesis is anti-runtime: most agent projects ship a custom orchestration layer; agency-agents ships recipes the existing harnesses already know how to execute. That makes it portable across ecosystems and durable across harness updates.

Why It Matters

By May 2026, the agent-team workflow is standard in indie dev and SMB consulting shops, but persona authoring is still a pain point — most teams crib their first round of personas from open-source repos rather than write them from scratch. agency-agents has captured that on-ramp role with 92K stars and a +828/day pace that suggests steady, non-viral adoption.

The pattern fits the broader 2026 thesis that the harness is the product and the personas are the variable: when the substrate is stable (Claude Code / Cursor / Codex), the value moves to who can write the best persona library. agency-agents is currently the largest such library by GitHub stars.

Use Cases

  • First-time multi-agent setup — replace “I should write personas eventually” with git clone and a one-line installer.
  • Consulting agencies — ship the same persona stack to every client engagement.
  • Indie shops — get a credible PM and a credible designer without hiring either.
  • Research teams — researcher and analyst personas with academic-citation guardrails.

Pairs With

  • TradingAgents — domain-specific multi-agent framework that complements general-purpose personas.
  • jcode — the harness; agency-agents provides the personas to run inside it.
  • Skills — for skill-based agents that can be composed alongside personas.

Verdict

If you’re standing up a multi-agent workflow and you don’t yet have strong opinions about persona design, agency-agents is the best starting library on GitHub. Treat it as a baseline to fork — the value isn’t that any single persona is perfect, it’s that you save the cold-start week of authoring them from scratch. Once you have your own house style, the markdown files are easy to mutate.

Similar Agents