AgentSkillsCN

Simple Commands

简单指令

SKILL.md

Simple Commands

The Rule

Run one command at a time. Look at the full output. Then decide what to do next.

Why

Chaining commands and truncating output looks efficient but causes problems:

bash
# Bad: chains commands, truncates output
npm test && npm run build 2>&1 | head -50

# Good: one command, full output
npm test

The Failure Mode

When you chain and truncate:

  1. Error might be on line 51 (past your head -50)
  2. You don't see it
  3. You run again, same result
  4. Human watches you loop while blind to the actual error

This is stochastic noise - doing plausible things that don't help.

The False Economy

"But chaining saves context!"

No. Re-running blind commands wastes context. One clear command with full output uses less context than three truncated attempts that miss the error.

When Chaining IS Okay

Sequential dependencies where you need all to succeed:

bash
# Fine: git operations that depend on each other
git add . && git commit -m "message" && git push

The key difference: you're not truncating, and failure at any step stops the chain.

Patterns to Avoid

bash
# Truncating build output
npm run build | head -50

# Hiding stderr
npm test 2>/dev/null

# Chaining unrelated commands
npm test && npm run lint && npm run build | tail -20

The Mindset

Each command is a question. Read the full answer before asking the next question.