Lobster Workflow

Use this when

• Starting any development task with LOBSTER
• Planning an implementation with /plan
• Coordinating parallel work with /team
• Running a quality review loop
• Needing to understand how LOBSTER's features work together

Overview

LOBSTER is an all-in-one AI development platform that turns your coding assistant into a coordinated development team. It combines smart context, implementation planning, multi-agent coordination, persistent memory, and cost tracking into a unified workflow.

Workflow Options

Option A: Quick Build (single agent)

1. Context auto-injects relevant files (automatic)
2. Use the coder agent to implement directly
3. Store learnings in memory when done

Option B: Planned Build (structured)

1. Use /plan to create an implementation plan
2. Review steps, dependencies, and risks
3. Work through steps in order, updating status as you go
4. Store learnings in memory when done

Option C: Team Build (parallel, multi-agent)

1. Use /plan to create an implementation plan (optional)
2. Use /team to decompose into subtasks and assign agents
3. Work through subtasks with specialized agents (coder, tester, reviewer, architect)
4. Track progress with team_status
5. Store learnings in memory when done

Option D: Review Loop (iterative quality)

1. Use review_loop for iterative code-review-test cycles
2. Coder generates code, reviewer reviews, tester tests
3. Fix issues and repeat until PASS

Full Workflow

Step 1: Context (automatic)

LOBSTER auto-injects relevant files, tech stack, and recent git activity into every message via <lobster-auto-context>. No action needed.

For manual context search:

find_relevant "description of what you want to build"

Step 2: Check Memory

Search for relevant past decisions and patterns:

memory_search "keywords related to your task"
memory_retrieve category:"architecture"

Step 3: Set Budget (Optional)

cost_budget budget_usd:5.00 alert_threshold:0.8

Step 4: Plan (Optional)

implementation_plan task:"Build a user authentication system" analyze_depth:"deep"

Step 5: Build

Choose your approach:

• Solo: Switch to coder agent and implement
• Team: Use team_coordinate to assign subtasks across agents
• Review loop: Use review_loop for iterative quality cycles

Step 6: Track Progress

plan_status          # Check plan progress
team_status          # Check team progress
review_status        # Check review loop progress

Step 7: Store Learnings

memory_store category:"pattern" title:"JWT auth pattern" content:"We use..." tags:["jwt", "auth"]

Step 8: Check Costs

cost_summary

Agent Reference

Quick checklist

• Let auto-context provide relevant files (or search manually)
• Check memory for past patterns and decisions
• Set budget if tracking costs
• Choose workflow: quick build, planned build, team build, or review loop
• Track progress as you work
• Store new learnings in memory
• Check cost summary when done

Converted and distributed by TomeVault — claim your Tome and manage your conversions.