davinci-resolve-mcp: Instructions file for Claude Code

CLAUDE.sessions.md

This file provides collaborative guidance and philosophy when using the Claude Code Sessions system.

Collaboration Philosophy

Core Principles:

• Investigate patterns - Look for existing examples, understand established conventions, don't reinvent what already exists
• Confirm approach - Explain your reasoning, show what you found in the codebase, get consensus before proceeding
• State your case if you disagree - Present multiple viewpoints when architectural decisions have trade-offs
• When working on highly standardized tasks: Provide SOTA (State of the Art) best practices
• When working on paradigm-breaking approaches: Generate "opinion" through rigorous deductive reasoning from available evidence

Task Management

Best Practices

• One task at a time (check .claude/state/current_task.json)
• Update work logs as you progress
• Mark todos as completed immediately after finishing

Quick State Checks

cat .claude/state/current_task.json  # Shows current task
git branch --show-current             # Current branch/task

current_task.json Format

ALWAYS use this exact format for .claude/state/current_task.json:

{
  "task": "task-name",        // Just the task name, NO path, NO .md extension
  "branch": "feature/branch", // Git branch (NOT "branch_name")
  "services": ["service1"],   // Array of affected services/modules
  "updated": "2025-08-27"     // Current date in YYYY-MM-DD format
}

Common mistakes to avoid:

• ❌ Using "task_file" instead of "task"
• ❌ Using "branch_name" instead of "branch"
• ❌ Including path like "tasks/m-task.md"
• ❌ Including .md file extension

Using Specialized Agents

You have specialized subagents for heavy lifting. Each operates in its own context window and returns structured results.

Prompting Agents

Agent descriptions will contain instructions for invocation and prompting. In general, it is safer to issue lightweight prompts. You should only expand/explain in your Task call prompt insofar as your instructions for the agent are special/requested by the user, divergent from the normal agent use case, or mandated by the agent's description. Otherwise, assume that the agent will have all the context and instruction they need.