research-domains

name	research-domains
description	Periodically research best practices across software, management, investment, and hobbies. Proposes improvements to this repository.
user-invocable	true
argument-hint	[domain or 'all']

Research Domains Skill

Purpose

Systematically research best practices across multiple domains and propose concrete improvements to the repository's skills, agents, rules, and cookbooks.

Argument Parsing

$ARGUMENTS specifies the domain(s) to research:

• software / management / investment / hobbies — single domain
• all or omitted — all domains

Domain-Specific Research Priorities

Management

In addition to general engineering management practices, prioritize organizational insights published by leading AI companies (Anthropic, OpenAI, DeepMind, etc.). These companies actively share practices on:

• Team structure and scaling (research ↔ engineering ↔ product)
• Decision-making frameworks at fast-scaling organizations
• Culture of writing and documentation (RFCs, design docs)
• Leadership principles and values frameworks

Launch a dedicated sub-agent for AI company organizational research alongside the general management researcher.

Workflow

Step 0: Load Previous Audit Proposals

Before launching new research, search the memory MCP for proposals saved by the quarterly-audit remote trigger:

1. Search the memory MCP via recall with query "Quarterly Audit Proposal" and top_k: 10
2. If results exist, include them as Prior Proposals in Step 2's output — these are pre-researched improvement suggestions that the user has not yet reviewed
3. If no results, skip this step silently

Step 1: Launch Domain Research

Launch the domain-researcher agent in the background via the Agent tool:

• Pass the target domain(s) from $ARGUMENTS
• The agent will query the memory MCP for user context and existing knowledge, and the web for current best practices

Step 2: Present Findings

When the agent returns, present findings in two sections:

Section A: Domain Summaries — key findings per domain, with source credibility tags

Section B: Improvement Proposals — numbered list of concrete changes to this repository. Each proposal includes:

1. What: the specific change (new skill, agent, rule, hook, or cookbook)
2. Where: target file path
3. Why: the best practice or gap that motivates the change
4. Domain: which domain this belongs to

Step 3: User Selection

Use AskUserQuestion (multiSelect) to let the user choose which proposals to implement.

Step 4: Plan and Implement

For each selected proposal:

1. Enter plan mode if the combined changes are non-trivial (2+ files)
2. Implement following existing patterns (skills in cookbooks/claude-code/files/skills/, agents in agents/, etc.)
3. Run ./bin/mitamae local linux.rb --dry-run to verify
4. Commit with descriptive message