Skill: tool-usage-discipline

Skill: tool-usage-discipline

What I do

I ensure the most efficient and accurate use of available tools. I prioritise high-context MCP tools and loaded skills over manual exploration, preventing reinventing the wheel and reducing context bloat.

When to use me

• Before starting any investigation or code change
• To decide whether to use a specific MCP tool or a manual bash command
• When facing a large codebase where manual navigation is inefficient
• To optimise token usage and session length

Core principles

1. Prioritise MCP — Use specialized tools (LSP, grep, glob) before generic ones (bash ls/​cat).
2. Consult skills first — Use loaded skills for domain expertise before seeking external information.
3. Avoid redundancy — Don't call a tool if you already have the information in your context.
4. Cache results — Store complex tool outputs (e.g. large grep results) in memory for the duration of the session.

Patterns & examples

Tool Selection Decision Matrix:

• Code Search: Use grep or ast_grep (fast, indexed) over manual find + cat.
• Navigation: Use lsp_goto_definition over manual searching.
• Verification: Use lsp_diagnostics before running a full build.
• Domain Knowledge: Use the skill() or vault-rag tools before web search.

Efficient Pattern:

• Inefficient: ls -R, cat file1, cat file2, grep "pattern" file1...
• Efficient: grep -r "pattern" followed by read on the most relevant match.

Anti-patterns to avoid

• ❌ Tool spam — Calling multiple tools to get information that a single, better tool could provide.
• ❌ Reinventing the tool — Writing a complex bash script when an MCP tool already handles that use case.
• ❌ Ignoring tool documentation — Using a tool sub-optimally because you haven't checked its parameters.

KB Reference

~/​vaults/​baphled/​3. Resources/​Knowledge Base/​AI Development System/​Skills/​Workflow-Orchestration/​Tool Usage Discipline.md

Related skills

• pre-action — Deciding on the best tool approach
• memory-keeper — Storing tool results to avoid repeat calls
• knowledge-base — Using specialized search tools
• token-efficiency — Optimising tool calls for token budget