Get up and running with Meridian in 5 minutes.
# Navigate to your project
cd /path/to/your/project
# Copy Meridian files
cp -r /path/to/meridian-opencode/.meridian ./
cp -r /path/to/meridian-opencode/.opencode ./
# Install plugin dependencies
cd .opencode/plugin && npm install && cd ../..Edit .meridian/config.yaml:
project_type: standard # Options: standard, hackathon, production
tdd_mode: false # Set to true for TDD workflowopencodeYou should see:
[Meridian] Project environment loaded. Core rules, guides, tasks, and memory are now active.
OpenCode will automatically:
- Read coding guides (120 baseline rules + any addons)
- Load memory entries
- Review task backlog
- Ask what you'd like to work on
In OpenCode:
I'd like to add user authentication to the app
OpenCode will:
- Discuss the approach
- Create a plan
- Ask for approval
You:
Sounds good, let's do it
OpenCode will:
- Use the
task-managertool - Create
TASK-001folder - Generate task files
- Update backlog
- Start implementation
During implementation, when OpenCode makes an architectural decision:
Document this decision using memory-curator:
We chose JWT tokens over sessions because we need stateless auth for our microservices architecture
OpenCode will:
- Use the
memory-curatortool - Add entry to
memory.jsonl - Tag appropriately
- Link to the current task
You're now using Meridian. Here's what happens automatically:
- ✅ Loads coding guides
- ✅ Reads memory entries
- ✅ Reviews tasks
- ✅ Ensures context continuity
- ✅ Enforces coding standards
- ✅ Creates structured tasks
- ✅ Documents decisions
- ✅ Tracks progress
- ✅ Updates task status
- ✅ Saves progress notes
- ✅ Runs tests/lint/build
- ✅ Ensures clean state
You: I want to add a shopping cart feature
OpenCode: [Creates plan with steps, acceptance criteria, risks]
You: Looks good, go ahead
OpenCode: [Uses task-manager to create TASK-002]
OpenCode: [Implements the feature]
OpenCode: [Updates task context with progress]
You: There's a bug in the checkout flow - users can't apply discount codes
OpenCode: [Investigates the issue]
OpenCode: [Creates a plan to fix it]
You: Approved
OpenCode: [Uses task-manager to create TASK-003]
OpenCode: [Writes failing test first if TDD mode enabled]
OpenCode: [Fixes the bug]
OpenCode: [Updates task and marks as done]
You: We need to decide between REST and GraphQL for our API
OpenCode: [Discusses trade-offs]
You: Let's go with GraphQL
OpenCode: [Uses memory-curator to document]
Entry: mem-0005
Decision: Use GraphQL for API layer
Problem: Need flexible querying for complex UI requirements
Alternatives: REST (rejected: too many endpoints), gRPC (rejected: browser support)
Trade-offs: More complex setup, but better DX and performance
Impact/Scope: All API routes in /api/graphql
Pattern: Use Apollo Server with code-first schema
# See all architectural decisions
jq -s '.[] | select(.tags | index("architecture"))' .meridian/memory.jsonl
# See decisions for a specific task
jq -s '.[] | select(.links | index("TASK-002"))' .meridian/memory.jsonl
# See recent entries
tail -5 .meridian/memory.jsonl | jq -s '.'# View backlog
cat .meridian/task-backlog.yaml
# Read a specific task
cat .meridian/tasks/TASK-001/TASK-001.yaml
cat .meridian/tasks/TASK-001/TASK-001-plan.md
cat .meridian/tasks/TASK-001/TASK-001-context.mdBest for: Most projects
project_type: standard- 120 baseline rules
- Balanced approach
- Structured workflow
- No extra constraints
Best for: Rapid prototyping, MVPs
project_type: hackathon- Relaxed standards
- Faster iteration
- Minimal testing
- Security still enforced
Example relaxations:
- Flat structure OK
- Client-side fetching allowed
- Larger components acceptable
- Skip dark mode/theming
- One E2E test sufficient
Best for: Production systems
project_type: production- Stricter standards
- Enhanced security
- Comprehensive testing
- Full observability
Example requirements:
- Design system mandatory
- Strict TypeScript config
- Rate limiting required
- Idempotency keys for mutations
- OpenTelemetry traces
Best for: Test-driven development
tdd_mode: true- Red → Green → Refactor
- Tests before code
- Works with any project type
Workflow:
- Write failing test
- Implement minimal code to pass
- Refactor while keeping tests green
- Let OpenCode create tasks
- Let OpenCode document decisions
- Review but don't micromanage
Only add entries that:
- Affect future features
- Define repeatable patterns
- Prevent future mistakes
Add timestamped notes for:
- Key decisions
- Blockers encountered
- Links to PRs/commits
- Lessons learned
OpenCode will prompt you to:
- Update task status
- Save progress notes
- Run tests/lint/build
- Document decisions
Edit these files to match your standards:
.meridian/CODE_GUIDE.md.meridian/prompts/agent-operating-manual.md.meridian/tasks/TASK-000-template/
cd .opencode/plugin
npm install# Verify files exist
ls -la .meridian/CODE_GUIDE.md
ls -la .meridian/config.yaml# Check plugin exports
grep "export const" .opencode/plugin/*.ts# Validate JSON
jq . .meridian/memory.jsonlNow that you're set up:
- Read the guides - Understand the coding standards
- Create a task - Practice the workflow
- Add memory - Document a decision
- Customize - Adapt to your team's needs
- Share - Get your team using Meridian
- README.md - Full documentation
- INSTALL.md - Detailed installation guide
- MIGRATION.md - Migrating from Claude Code
- .meridian/prompts/agent-operating-manual.md - Agent behavior
- .meridian/CODE_GUIDE.md - Coding standards
Questions? Check:
- This quick start guide
- The main README.md
- The installation guide
- OpenCode docs: https://opencode.ai/docs/plugins
Happy coding with Meridian! 🚀