Curated registry of 69 AI agents for OpenCode, distributed via a zero-dependency CLI and interactive TUI. Agents are .md files containing system prompts that configure specialized AI assistants.
Each agent follows an expert 4-section format: identity, decisions, examples, quality gate.
# Interactive TUI with intelligent suggestions (auto-detects TTY)
npx github:dmicheneau/opencode-template-agent
# Detect stack and suggest relevant agents
npx github:dmicheneau/opencode-template-agent install
# Quick CLI — install a pack in one command
npx github:dmicheneau/opencode-template-agent install --pack backendThe TUI is the easiest way to discover and install agents. It launches automatically when the terminal supports TTY.
npx github:dmicheneau/opencode-template-agent
# or explicitly:
npx github:dmicheneau/opencode-template-agent tuiWhat it does:
- Auto-detects TTY and launches the interactive interface
- Suggestion screen on startup: if a stack is detected in the current project, the most relevant agents appear pre-selected — press
Spaceto deselect,Enterto install,Bto browse the full catalog - Browse categories with tabs (
←→/Tab) - Navigate lists with
↑↓ - Select agents with
Space, install withEnter - Built-in search (
/), pack and category browsing
For automation or quick installation without a graphical interface.
Commands:
# Detect stack and suggest relevant agents
npx github:dmicheneau/opencode-template-agent install
# Install a specific agent
npx github:dmicheneau/opencode-template-agent install typescript-pro
# Install one or more packs (comma or space separated)
npx github:dmicheneau/opencode-template-agent install --pack backend
npx github:dmicheneau/opencode-template-agent install --pack backend,devops
# Install one or more categories
npx github:dmicheneau/opencode-template-agent install --category languages
npx github:dmicheneau/opencode-template-agent install --category languages,data-api
# Install all agents
npx github:dmicheneau/opencode-template-agent install --all
# List all agents by category
npx github:dmicheneau/opencode-template-agent list
# List available packs
npx github:dmicheneau/opencode-template-agent list --packs
# Search agents
npx github:dmicheneau/opencode-template-agent search docker
npx github:dmicheneau/opencode-template-agent search "machine learning"Options:
| Option | Description |
|---|---|
--force |
Overwrite existing files; with install (no args), re-suggests already-installed agents |
--dry-run |
Preview without writing to disk (also works with the auto-suggest flow) |
Note:
--packand--categoryare mutually exclusive.
Via bash script:
curl -fsSL https://raw.githubusercontent.com/dmicheneau/opencode-template-agent/main/install.sh | bashOptions: --copy, --global, --dir PATH, --force, --dry-run, --uninstall.
From a local clone:
git clone https://github.com/dmicheneau/opencode-template-agent.git ~/.opencode-agents
echo 'export OPENCODE_CONFIG_DIR=~/.opencode-agents' >> ~/.zshrc
source ~/.zshrcAgents are installed into .opencode/agents/. Node.js 20+ required for modes 1 and 2.
The plugin exposes the 69-agent registry as LLM-callable tools inside OpenCode sessions. Current version: discovery-only (v1) — 4 read-only tools, no write operations.
Global (recommended — available in all OpenCode sessions):
Add to ~/.config/opencode/opencode.json:
{
"plugin": ["github:dmicheneau/opencode-template-agent"]
}Per-project (available only in that project):
Add to .opencode/opencode.json at the project root:
{
"plugin": ["github:dmicheneau/opencode-template-agent"]
}| Tool | Description |
|---|---|
search_agents |
Search by name, description, category, or tags |
list_agents |
List all agents grouped by category, with optional category or pack filters |
get_agent |
Full details for a specific agent (description, install mode, install state, typo suggestions) |
check_health |
Registry health report (installed, outdated, missing, unknown agents, file integrity) |
v1.1 —
Oc*-prefixed types to prevent namespace collisions,readonlymode for lock file operations, and absolute path sanitization in error messages.
> Search for a TypeScript agent
> List agents in the security category
> Check registry health
plugin/
index.ts # entry point (exported as OpenCode Plugin)
tools.ts # 4 tool definitions
types.d.ts # type declarations
Requirements: OpenCode + Bun (TypeScript is transpiled natively, no build step needed).
Full documentation:
docs/architecture.md
The diagram below shows the overall system architecture: the CLI entry point, TUI modules, agent registry, and upstream monitoring scripts.
flowchart TB
User["User"]
subgraph CLI["bin/cli.mjs — CLI Entry Point"]
Parse["Argument parsing<br/>(install, list, search, tui)"]
end
subgraph TUI["Interactive TUI"]
Index["index.mjs<br/>Orchestrator<br/>(lifecycle, main loop,<br/>signals)"]
Screen["screen.mjs<br/>Terminal I/O<br/>(raw mode, flush,<br/>resize, onInput)"]
Input["input.mjs<br/>Keystroke parser<br/>(raw bytes → ~20 Actions)"]
State["state.mjs<br/>State machine<br/>(browse, search, confirm,<br/>installing, pack_detail,<br/>done, quit)"]
Renderer["renderer.mjs<br/>Frame builder<br/>(state → ANSI string)"]
Ansi["ansi.mjs<br/>ANSI codes, colors,<br/>box drawing, palettes<br/>(catColor, tabColor)"]
end
subgraph Data["Data Layer"]
Registry["registry.mjs<br/>Manifest loader<br/>(validation, getAgent,<br/>getCategory, searchAgents,<br/>resolvePackAgents)"]
Manifest["manifest.json<br/>69 agents | 10 categories<br/>15 packs"]
Installer["installer.mjs<br/>GitHub raw download<br/>→ .opencode/agents/"]
end
subgraph Sync["Upstream monitoring (manual dispatch)"]
Upstream["davila7/claude-code-templates<br/>(upstream repo)"]
SyncScript["sync-agents.py<br/>(1200 lines, fetch,<br/>tools→permission conversion,<br/>CURATED + EXTENDED agents)"]
SyncCommon["sync_common.py<br/>(HTTP, ETag cache,<br/>frontmatter, validation)"]
UpdateManifest["update-manifest.py<br/>(manifest merge,<br/>NEEDS_REVIEW prefix)"]
GHA["GitHub Actions<br/>(workflow_dispatch only,<br/>CI: test + lint + validate)"]
end
LocalDir[".opencode/agents/<br/>Installed agents"]
User --> CLI
Parse -->|"tui"| Index
Parse -->|"install"| Installer
Parse -->|"list / search"| Registry
Index --> Screen
Screen --> Input
Input --> State
State --> Renderer
Renderer --> Ansi
Ansi -->|"ANSI frames"| Screen
Index --> Registry
Index --> Installer
Registry --> Manifest
Installer -->|"downloads"| LocalDir
Upstream --> SyncScript
SyncScript --> SyncCommon
SyncScript --> UpdateManifest
UpdateManifest --> Manifest
GHA -->|"orchestrates"| SyncScript
GHA -->|"orchestrates"| UpdateManifest
classDef entrypoint fill:#4a90d9,stroke:#2c5f8a,color:#fff
classDef tui fill:#6ab04c,stroke:#3d7a28,color:#fff
classDef data fill:#f0932b,stroke:#c0741e,color:#fff
classDef sync fill:#9b59b6,stroke:#6c3483,color:#fff
classDef storage fill:#e74c3c,stroke:#a93226,color:#fff
classDef user fill:#34495e,stroke:#1c2833,color:#fff
class User user
class Parse entrypoint
class Index,Screen,Input,State,Renderer,Ansi tui
class Registry,Manifest,Installer data
class Upstream,SyncScript,SyncCommon,UpdateManifest,GHA sync
class LocalDir storage
Two additional diagrams are available in docs/architecture.md:
- TUI user flow — complete state machine (suggest, browse, search, confirm, installing, done)
- Agent update pipeline — upstream monitoring scripts and discovery/evaluation workflow (manual dispatch only)
69 agents across 10 categories, invocable via @category/name.
| Category | Agents | Description |
|---|---|---|
| 💻 Languages | 11 | TypeScript, Python, Go, Rust, Java, C#, PHP, Kotlin, C++, Rails, Swift |
| 🤖 AI | 9 | AI engineering, data science, ML, MLOps, LLM, prompts, research, data engineering, data analysis |
| 🌐 Web | 9 | React, Next.js, Vue, Angular, mobile, UI design, UI analysis, accessibility, fullstack |
| 🗄️ Data & API | 6 | API architecture, GraphQL, databases, PostgreSQL, Redis, SQL |
| ⚙️ DevOps | 9 | Docker, Kubernetes, Terraform, AWS, CI/CD, Linux, platform, SRE, incident response |
| 🛠️ DevTools | 8 | Code review, debugging, performance, refactoring, testing, orchestration, microservices, QA, legacy modernization |
| 🔒 Security | 5 | Security audit, penetration testing, smart contracts, security engineering, compliance |
| 🔌 MCP | 2 | MCP development, MCP security audit |
| 📊 Business | 6 | Product management, project management, PRD, Scrum, UX research, business analysis |
| 📝 Docs | 4 | Technical documentation, API docs, writing, diagrams |
Each agent is automatically scored by scripts/quality_scorer.py across 8 dimensions (score 1–5 each):
| Dimension | What is measured | 5/5 |
|---|---|---|
frontmatter |
Presence of description, mode, permission |
All 3 fields present |
identity |
Identity paragraph between the frontmatter and the first ## |
50–300 words |
decisions |
## Decisions section with IF/THEN trees |
≥ 5 rules |
examples |
## Examples section with code blocks |
≥ 3 examples |
quality_gate |
## Quality Gate section with validation criteria |
≥ 5 criteria |
conciseness |
Line count (70–120 ideal) and filler phrase ratio | 70–120 lines, ≤ 3% filler |
no_banned_sections |
No legacy sections (Workflow, Tools, Anti-patterns, Collaboration) | 0 banned sections |
version_pinning |
Version and year references in the identity | Version + year present |
Pass threshold: average ≥ 3.5 AND no dimension < 2
Labels: Excellent (≥ 4.5) · Good (≥ 3.5) · Needs improvement (≥ 2.5) · Poor (< 2.5)
69 agents · Average score: 4.59/5 · 100% pass rate · 49 Excellent, 20 Good
Estimated token cost:
size in bytes / 4(approximation for English + code content).
| Category | Agent | Score | Label | ~Tokens | Lines |
|---|---|---|---|---|---|
| ai | ai-engineer | 4.75 | Excellent | 1,164 | 113 |
| ai | data-analyst | 4.75 | Excellent | 1,088 | 102 |
| ai | data-engineer | 4.75 | Excellent | 1,180 | 106 |
| ai | data-scientist | 4.75 | Excellent | 1,218 | 108 |
| ai | llm-architect | 4.88 | Excellent | 1,353 | 125 |
| ai | ml-engineer | 4.75 | Excellent | 1,170 | 108 |
| ai | mlops-engineer | 4.75 | Excellent | 1,205 | 125 |
| ai | prompt-engineer | 4.75 | Excellent | 1,386 | 121 |
| ai | search-specialist | 4.62 | Excellent | 1,316 | 114 |
| business | business-analyst | 4.62 | Excellent | 1,260 | 104 |
| business | prd | 4.25 | Good | 1,407 | 74 |
| business | product-manager | 4.25 | Good | 1,043 | 85 |
| business | project-manager | 4.38 | Good | 1,174 | 89 |
| business | scrum-master | 4.25 | Good | 1,265 | 98 |
| business | ux-researcher | 4.25 | Good | 1,446 | 116 |
| data-api | api-architect | 4.75 | Excellent | 1,351 | 128 |
| data-api | database-architect | 4.50 | Excellent | 1,265 | 113 |
| data-api | graphql-architect | 4.88 | Excellent | 1,249 | 128 |
| data-api | postgres-pro | 4.50 | Excellent | 1,208 | 119 |
| data-api | redis-specialist | 4.88 | Excellent | 1,243 | 122 |
| data-api | sql-pro | 4.50 | Excellent | 2,136 | 165 |
| devops | aws-specialist | 4.88 | Excellent | 1,087 | 123 |
| devops | ci-cd-engineer | 4.62 | Excellent | 1,103 | 118 |
| devops | docker-specialist | 4.62 | Excellent | 1,089 | 130 |
| devops | incident-responder | 4.25 | Good | 2,112 | 182 |
| devops | kubernetes-specialist | 4.88 | Excellent | 1,111 | 136 |
| devops | linux-admin | 4.62 | Excellent | 1,044 | 127 |
| devops | platform-engineer | 4.88 | Excellent | 1,019 | 118 |
| devops | sre-engineer | 4.38 | Good | 1,157 | 122 |
| devops | terraform-specialist | 4.88 | Excellent | 1,242 | 139 |
| devtools | code-reviewer | 4.25 | Good | 1,214 | 110 |
| devtools | debugger | 4.25 | Good | 1,369 | 122 |
| devtools | legacy-modernizer | 4.25 | Good | 2,661 | 220 |
| devtools | microservices-architect | 4.50 | Excellent | 1,230 | 150 |
| devtools | performance-engineer | 4.25 | Good | 1,226 | 119 |
| devtools | qa-expert | 4.25 | Good | 1,267 | 123 |
| devtools | refactoring-specialist | 4.50 | Excellent | 1,753 | 186 |
| devtools | test-automator | 4.50 | Excellent | 1,546 | 161 |
| docs | api-documenter | 4.62 | Excellent | 1,195 | 118 |
| docs | diagram-architect | 4.62 | Excellent | 1,173 | 111 |
| docs | documentation-engineer | 4.38 | Good | 1,050 | 105 |
| docs | technical-writer | 4.25 | Good | 1,097 | 120 |
| languages | cpp-pro | 4.62 | Excellent | 1,197 | 141 |
| languages | csharp-developer | 4.62 | Excellent | 1,162 | 133 |
| languages | golang-pro | 4.88 | Excellent | 1,101 | 129 |
| languages | java-architect | 4.88 | Excellent | 1,383 | 138 |
| languages | kotlin-specialist | 4.88 | Excellent | 1,191 | 121 |
| languages | php-pro | 4.88 | Excellent | 1,193 | 138 |
| languages | python-pro | 4.88 | Excellent | 1,139 | 139 |
| languages | rails-expert | 4.88 | Excellent | 1,232 | 139 |
| languages | rust-pro | 4.88 | Excellent | 1,211 | 139 |
| languages | swift-expert | 4.88 | Excellent | 1,201 | 138 |
| languages | typescript-pro | 4.88 | Excellent | 1,268 | 132 |
| mcp | mcp-developer | 4.88 | Excellent | 1,384 | 125 |
| mcp | mcp-security-auditor | 4.12 | Good | 1,289 | 87 |
| security | compliance-auditor | 4.75 | Excellent | 1,805 | 107 |
| security | penetration-tester | 4.62 | Excellent | 1,828 | 137 |
| security | security-auditor | 4.25 | Good | 1,633 | 104 |
| security | security-engineer | 4.25 | Good | 1,111 | 109 |
| security | smart-contract-auditor | 4.75 | Excellent | 2,269 | 126 |
| web | accessibility | 4.50 | Excellent | 1,273 | 107 |
| web | angular-architect | 4.25 | Good | 1,282 | 125 |
| web | fullstack-developer | 4.62 | Excellent | 1,034 | 103 |
| web | mobile-developer | 4.50 | Excellent | 1,223 | 125 |
| web | nextjs-developer | 4.25 | Good | 1,201 | 126 |
| web | react-specialist | 4.88 | Excellent | 994 | 104 |
| web | screenshot-ui-analyzer | 4.25 | Good | 1,380 | 99 |
| web | ui-designer | 4.62 | Excellent | 1,131 | 103 |
| web | vue-expert | 4.88 | Excellent | 1,094 | 104 |
15 predefined packs for installing coherent groups of agents in a single command.
| Pack | Agents | Description |
|---|---|---|
backend |
postgres-pro, redis-specialist, database-architect, api-architect, python-pro, typescript-pro, debugger, test-automator | Backend stack |
frontend |
react-specialist, nextjs-developer, typescript-pro, ui-designer, performance-engineer, test-automator | Frontend stack |
devops |
docker-specialist, kubernetes-specialist, terraform-specialist, aws-specialist, ci-cd-engineer, linux-admin, platform-engineer, incident-responder | Infrastructure |
fullstack |
fullstack-developer, typescript-pro, react-specialist, nextjs-developer, postgres-pro, api-architect, debugger, test-automator, code-reviewer | Full stack |
ai |
ai-engineer, data-scientist, ml-engineer, llm-architect, prompt-engineer, search-specialist | AI & ML |
security |
security-auditor, penetration-tester, smart-contract-auditor, compliance-auditor | Security |
mcp |
mcp-developer, mcp-security-auditor | MCP servers |
quality |
code-reviewer, test-automator, debugger, performance-engineer, refactoring-specialist, legacy-modernizer | Code quality |
startup |
fullstack-developer, typescript-pro, nextjs-developer, postgres-pro, docker-specialist, product-manager, ui-designer, test-automator | Startup kit |
data-stack |
data-engineer, data-analyst, data-scientist, database-architect, postgres-pro, sql-pro | Data stack |
ml-to-production |
data-scientist, ml-engineer, mlops-engineer, llm-architect, docker-specialist, kubernetes-specialist | ML to production |
frontend-complete |
react-specialist, nextjs-developer, vue-expert, angular-architect, accessibility, ui-designer | Complete frontend |
ship-it-safely |
ci-cd-engineer, docker-specialist, kubernetes-specialist, sre-engineer, security-engineer, qa-expert | Safe deployment |
product-discovery |
product-manager, ux-researcher, business-analyst, prd, ui-designer | Product discovery |
architecture-docs |
microservices-architect, api-architect, database-architect, diagram-architect, documentation-engineer | Architecture & docs |
Every push or pull request to main triggers 4 parallel jobs:
| Job | Description |
|---|---|
| test | Python tests across 3 versions (3.10, 3.12, 3.13) |
| test-cli | Node.js tests across 3 versions (20, 22, 23) — CLI, TUI, lock |
| lint | Python/Node syntax, shellcheck, agent YAML frontmatter validation, manifest JSON validation |
| validate-agents | Verifies manifest consistency with actual files, detects deprecated fields |
Dependabot monitors the SHA pins of GitHub Actions used in workflows and automatically opens a PR each week when an update is available. All actions are pinned by SHA for security.
Agents come from multiple sources: the majority originates from aitmpl.com (claude-code-templates), some are original creations of this project (aws-specialist, ci-cd-engineer, docker-specialist, linux-admin, redis-specialist, mcp-security-auditor, prd, screenshot-ui-analyzer), and the awesome-claude-code-subagents catalog was used as a reference to identify missing agents. All agents are manually curated and go through an expert rewriting process before integration — the automated weekly sync has been disabled to ensure high quality.
Upstream agents (~133 available) follow a generic format (capability lists, fictional metrics). Project agents follow an expert 4-section format (identity, decisions, examples, quality gate). The quality gap (3-4/10 vs 8-9/10) makes automated import counterproductive.
- Discovery — list available upstream agents using the local script:
python3 scripts/sync-agents.py --list --tier=extended
- Evaluation — verify the agent fills a gap not covered by the existing 69 agents
- Upstream dry-run — trigger the workflow in discovery mode to fetch frontmatter and permissions without modifying the repo:
gh workflow run "Sync Agents" -f tier=core -f dry_run=true - Rewriting — rewrite the body using the project's template (Identity → Decisions → Examples → Quality Gate)
These scripts are for manual use only — there is no automated sync.
| Script | Description |
|---|---|
scripts/sync-agents.py |
Downloads and converts agents from the upstream repo |
scripts/update-manifest.py |
Merges sync manifest with the main manifest |
scripts/sync_common.py |
Shared HTTP utilities, ETag cache, frontmatter validation |
scripts/generate_readme_scores.py |
Regenerates quality score tables in both READMEs. --check mode for CI verification. |
The changelog is automatically generated from Git history using git-cliff, user-oriented with clear categories.
- Tag push — push a
v*tag (e.g.,git tag v8.0.0 && git push --tags) - Changelog generation — git-cliff analyzes commits since the last tag and generates a structured changelog
- GitHub Release — a release is automatically created with the changelog as the body
| Commit prefix | Changelog category |
|---|---|
feat |
✨ New features |
fix |
🐛 Bug fixes |
perf |
⚡ Performance |
docs |
📝 Documentation |
refactor |
♻️ Refactoring |
chore, ci, build, style, test |
🔧 Maintenance |
Commits with
BREAKING CHANGEare prefixed with BREAKING: in their respective category.
# Bump version in package.json, tag and push
npm version major # or minor, patch
git push --follow-tags
# Or manually
git tag v8.0.0
git push --tags814 tests (504 JS + 310 Python).
# All JS tests (CLI + TUI)
node --test tests/cli.test.mjs tests/tui.test.mjs tests/lock.test.mjs
# All Python tests
python3 tests/run_tests.py
# Specific tests
python3 -m pytest tests/test_agents.py -v
python3 -m pytest tests/test_sync_script.py -v
python3 -m pytest tests/test_update_manifest.py -vContributions are welcome! See the open issues.
- Create or edit the file in
agents/<category>/<name>.md - Follow the 4-section template: Identity, Decisions, Examples, Quality Gate
- Regenerate the README scores:
python3 scripts/generate_readme_scores.py
- Commit the agent files and the updated READMEs together
- CI will automatically verify that scores are up to date
Tip: run
python3 scripts/quality_scorer.py agents/<category>/<name>.mdto see the detailed score before updating the READMEs.
Each agent must meet the quality threshold: average score ≥ 3.5 and no dimension < 2. See the Agent quality section for dimension details.
Agents primarily sourced from aitmpl.com (claude-code-templates), with original project creations and awesome-claude-code-subagents as reference. All manually rewritten and curated.
MIT