MI

maslennikov-ig/template-bridge

Deployment & DevOps
73 stars Quality 65 Trend 18MIT

Unified Claude Code workflow: Superpowers + Beads + Templates. 413+ specialist agents on-demand.

Overview

Unified development workflow for Claude Code: . Bridges three powerful Claude Code plugins into a single disciplined workflow with on-demand access to 413+ specialist agent templates. Both are optional — the workflow skill adapts, but you get the full power with all three. No production code without a failing test first. Ever. Always invoke superpowers:systematic-debugging BEFORE any fix: 1. — read errors, reproduce, check recent changes 2. — find working examples, compare 3. — single-variable test 4. — with failing test first 5. after 3 failed attempts → question architecture 1. Identify the proving command 2. Run it fresh 3. Read complete output + exit code 4. Verify it supports the claim 5. Only then claim done When a task needs a specialist not in current skills:

README

Template Bridge Plugin

Unified development workflow for Claude Code: Superpowers + Beads + Templates.

Bridges three powerful Claude Code plugins into a single disciplined workflow with on-demand access to 413+ specialist agent templates.

Prerequisites

Plugin Install Purpose
Superpowers /plugin install superpowers TDD, brainstorming, debugging, code review
Beads /plugin install beads Task tracking across sessions

Both are optional — the workflow skill adapts, but you get the full power with all three.

What’s Inside

Component Type Purpose
unified-workflow Skill How to work: beads task → brainstorm → plan → TDD → review → finish
template-catalog Skill 413+ specialist agents on-demand from claude-code-templates
browse-templates Command /browse-templates — manual template search

The Stack

┌─────────────────────────────────────────────────────┐
│                   YOUR TASK                          │
├──────────┬──────────────────┬───────────────────────┤
│  Beads   │   Superpowers    │   Template Bridge     │
│  WHAT    │   HOW            │   WHO                 │
│          │                  │                       │
│ bd create│ brainstorming    │ template-catalog      │
│ bd ready │ writing-plans    │ 413+ specialist agents│
│ bd close │ TDD (red-green)  │ on-demand install     │
│          │ debugging        │                       │
│          │ code-review      │                       │
│          │ verification     │                       │
└──────────┴──────────────────┴───────────────────────┘

The Flow

Every task follows this sequence:

1. TASK        bd create -t feature "..."  or  bd update  --status in_progress
                │
2. BRAINSTORM   superpowers:brainstorming → design before code
                │
3. PLAN         superpowers:writing-plans → 2-5 min tasks
                │
4. ISOLATE      superpowers:using-git-worktrees (non-trivial work)
                │
5. IMPLEMENT    For each task:
                ├─ RED:    write ONE failing test
                ├─ VERIFY: confirm it fails for the right reason
                ├─ GREEN:  minimal code to pass
                ├─ VERIFY: all tests pass
                ├─ REFACTOR: clean up
                └─ COMMIT
                │
6. REVIEW       superpowers:requesting-code-review
                │
7. VERIFY       superpowers:verification-before-completion
                │
8. FINISH       superpowers:finishing-a-development-branch
                │
9. CLOSE        bd close  --reason "Done"

TDD: Red-Green-Refactor

No production code without a failing test first. Ever.

RED ──────► VERIFY RED ──────► GREEN ──────► VERIFY GREEN ──────► REFACTOR ──────► COMMIT
write test   fails for        minimal        all tests           clean up          after each
             right reason     code           pass                                 green

Debugging

Always invoke superpowers:systematic-debugging BEFORE any fix:

  1. Root cause — read errors, reproduce, check recent changes
  2. Pattern analysis — find working examples, compare
  3. Hypothesis — single-variable test
  4. Fix — with failing test first
  5. Stop after 3 failed attempts → question architecture

Verification Gate

Before claiming anything done:

  1. Identify the proving command
  2. Run it fresh
  3. Read complete output + exit code
  4. Verify it supports the claim
  5. Only then claim done

“Evidence before claims, always.”

External Templates

When a task needs a specialist not in current skills:

# Direct install (always fresh from repo)
npx claude-code-templates@latest --agent security/security-auditor --yes

# Interactive browser
npx claude-code-templates@latest

Categories (26)

Category Top Agents
ai-specialists prompt-engineer, llm-architect, model-evaluator
api-graphql api-architect, graphql-specialist
blockchain-web3 smart-contract-auditor, blockchain-developer
data-ai data-engineer, ml-engineer, data-scientist
database postgres-specialist, redis-expert
devops-infrastructure kubernetes-specialist, terraform-expert, cloud-architect
development-team backend-architect, frontend-developer, mobile-developer
development-tools code-optimizer, dependency-manager
game-development unity-developer, godot-specialist
mcp-dev-team mcp-server-architect, mcp-developer
modernization legacy-modernizer, architecture-migrator
performance-testing performance-engineer, web-vitals-optimizer
programming-languages rust-expert, go-specialist, python-architect
security security-auditor, penetration-tester
ui-analysis screenshot-analyzer, accessibility-auditor
web-tools seo-analyzer, accessibility-specialist

Full list: 26 categories, 413+ agents

Refreshing the catalog

The category table in skills/template-catalog/SKILL.md is auto-generated from the upstream cli-tool/components/agents tree (the same source npx --agent installs from — not the incomplete docs/api/agents.json). To resync counts and names after upstream changes:

node scripts/sync-catalog.mjs
# Optional: raise the GitHub API rate limit (60/h anon → 5000/h)
GITHUB_TOKEN=ghp_xxx node scripts/sync-catalog.mjs

Requires curl and Node 18+. The script rewrites only the block between the / markers plus the agent/category counts. It exits non-zero and leaves the file untouched if the upstream layout changed or zero agents were found — no silently-wrong table.

Skill Priority

  1. Process — brainstorming, debugging, verification (always first)
  2. Implementation — TDD, code-review, frontend-design (during work)
  3. Templates — only when no existing skill covers the need

Anti-Patterns

  • Skip brainstorming → “just code it”
  • Write code before failing test
  • Claim “fixed” without running verification
  • Work without beads task
  • Multiple fixes simultaneously
  • Trust subagent reports without own verification
  • Qualifier language (“should work”, “probably fixed”)

Installation

claude plugin marketplace add maslennikov-ig/template-bridge
claude plugin install template-bridge@template-bridge-marketplace

From local clone

git clone https://github.com/maslennikov-ig/template-bridge.git
claude plugin marketplace add /path/to/template-bridge
claude plugin install template-bridge@template-bridge-marketplace

Three layers of protection

The plugin uses three layers to ensure Claude never forgets the workflow, even in a fresh context:

Layer Mechanism How it works
1. CLAUDE.md System prompt (highest priority) Rules loaded before anything else, every session
2. SessionStart hook Echo reminder after bd prime Explicit reminder in context on session start
3. Skill description Plugin skill listing unified-workflow trigger visible in skill list

Layer 1: Copy CLAUDE.md to global config

The repo includes a ready-to-use CLAUDE.md. Copy its contents to your global config:

# Append to your global CLAUDE.md (create if doesn't exist)
cat /path/to/template-bridge/CLAUDE.md >> ~/.claude/CLAUDE.md

Or copy manually — the file is short (the full 9-step workflow + 4 rules).

The repo includes settings.example.json with two hooks:

  • SessionStart — bd prime (loads Beads context) + workflow reminder
  • PreCompact — bd prime (re-injects Beads context before compaction so tasks survive)

Merge into your ~/.claude/settings.json:

# View the example first
cat /path/to/template-bridge/settings.example.json

# Then manually merge the "hooks" section into your ~/.claude/settings.json

Or copy the hooks block directly:

{
  "hooks": {
    "PreCompact": [
      {
        "matcher": "",
        "hooks": [
          { "type": "command", "command": "bd prime" }
        ]
      }
    ],
    "SessionStart": [
      {
        "matcher": "",
        "hooks": [
          { "type": "command", "command": "bd prime" },
          { "type": "command", "command": "echo 'WORKFLOW REMINDER: Invoke template-bridge:unified-workflow before any task. Flow: bd create → brainstorm → plan → TDD → review → verify → finish → bd close'" }
        ]
      }
    ]
  }
}

Layer 3: Automatic

The skill description is always visible in Claude’s skill listing — no setup needed.

Updating

claude plugin install template-bridge@template-bridge-marketplace

Plugin re-reads from the source on reinstall.

Contributing

  1. Fork this repo
  2. Edit skills or add new ones
  3. Submit a PR

Plugin structure follows the official Claude Code plugin spec.

Credits

  • Superpowers by Jesse Vincent — the TDD/brainstorming/debugging methodology
  • Beads by Steve Yegge — task tracking for AI workflows
  • claude-code-templates by Daniel Avila — 413+ specialist agent templates

License

MIT

View the full README on GitHub

Recommended Tools

Try a different keyword or remove a filter.

Install

npx skillfish add maslennikov-ig/template-bridge
New to MCP? Start with the mcpskill.org guide →