Agency Agents
github.com/msitarzewski/agency-agents- Category
- AI Agents
- Rank
- No. 145Tools index
- Type
- APP
- Builder
- msitarzewski
- GitHub
- 152.4k stars
- Date
About
A ready-made AI agency — a roster of specialized agents (frontend, marketing, community, and more) you can deploy on demand.
What it does
Agency Agents is a library of detailed Markdown role specifications for AI coding tools. Each specification defines a persona, operating rules, workflows, deliverables, and success measures. An installer converts or copies selected agents into formats understood by supported assistants and editors.
Why it's ranked here
The project pairs unusually specific prompt content with practical distribution tooling. Its strongest distinction is portability: one source collection can target many assistants through per-agent files, combined rosters, or a plugin. The examples and contribution template also make its intended operating model concrete.
What's good
Agents contain more than role labels. The supplied GIS and design examples prescribe checks, phased processes, output templates, tool knowledge, and explicit boundaries. Users can install individual agents or whole divisions. Tool definitions centralize destination, scope, rendering format, detection, and installation mechanism.
Tradeoffs
These are prompt definitions, not independent workers or a hosted orchestration service. Quality and factual accuracy still depend on the model running them. Some targets require conversion before installation, while others accept only project-scoped rosters. OpenCode may silently omit agents beyond its runtime limit, so selective installation matters.
How to use it well
Use it when your existing coding assistant needs a repeatable specialist stance, concrete deliverables, or a disciplined checklist. Start with one agent or division, adapt its rules to your project, and inspect generated files before installation. Choose another system if you need model hosting, execution infrastructure, or guaranteed cross-agent coordination.
Technical notes+
tools.json defines 16 target contracts, including render format, install kind, scope, detection directories, version command, and destination templates. divisions.json declares 17 source divisions. scripts/lib.sh provides Bash 3.2-compatible frontmatter extraction, slug generation, terminal capability checks, and interactive TUI primitives; its YAML handling reads simple first-match fields rather than using a full parser. integrations/README.md documents conversion and installation behavior across supported tools. SECURITY.md distinguishes non-executable Markdown prompts from executable shell scripts and advises reviewing the scripts before use. CONTRIBUTING.md specifies the common agent structure and explains that generated integration output is not committed.
Observed
- License
- MIT License
- Primary language
- Markdown agent definitions with Bash automation scripts
- CLI surface
- Shell-based converter and interactive installer, with tool, division, agent, listing, and dry-run selection
- Install formats
- Per-agent files, combined roster files, and a plugin artifact
- Supported tool contracts
- 16 targets declared in the central tool catalog
- Platform support
- The linked native desktop app supports macOS, Linux, and Windows
- Repository structure
- 17 agent divisions declared in the central division catalog
Read from README.md, LICENSE, tools.json, SECURITY.md, divisions.json, CONTRIBUTING.md, CONTRIBUTING_zh-CN.md, scripts/lib.sh, examples/README.md, gis/gis-analyst.md, .github/FUNDING.yml, gis/gis-qa-engineer.md, integrations/README.md, gis/gis-bim-specialist.md, design/design-ui-designer.md.
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.