Rules files are loaded often, so every unnecessary paragraph uses context on many requests. Long files also make important guidance harder to notice. Conflicting or overly detailed rules can cause the assistant to follow the wrong instruction, or to spend time reconciling advice that does not apply.
Keep only stable, high-impact facts that change how work should be done.
Use clear commands, short examples, and links to deeper documentation.
Move subsystem guidance into scoped files when the tool supports them.
Remove rules that a formatter, type checker, or test already enforces.
Review the file after repeated agent mistakes and after architecture changes.
Lean rules reduce cost, improve attention, and are easier for people to keep accurate.
Rewriting in plainer words…
This answer doesn't lend itself to a diagram - it reads best . No credits were charged.
Why delete a rule that a linter or type checker already enforces? Isn't the extra reminder harmless?The tool enforces it reliably, the rule costs context on every request, and a restated rule can drift away from the tool's real config.
Rules files load on many requests, so every extra paragraph costs context again and again, and a long file buries the guidance that matters.
Conflicting or overly detailed rules make it worse, because the assistant may follow the wrong one or spend effort reconciling advice that doesn't apply.
So I keep only stable, high-impact facts that change how the work gets done, written as clear commands with short examples and links to deeper docs.
Subsystem guidance goes into scoped files where the tool supports them, and I delete any rule a formatter, type checker or test already enforces.
The payoff is lower cost, better attention on the rules that remain, and a file people can keep accurate.
The catch is that anything moved out of the file only helps if the agent finds it, so I review the file after repeated agent mistakes and after architecture changes.
Weak answers to avoid
Only talks about token costCost is half of it. Long or conflicting rules also bury the important ones and can make the agent follow the wrong instruction. Cover attention and conflicts too.
Says bigger context windows solved thisA bigger window doesn't remove the per-request cost or the attention problem. More text fits, but it still competes with the rules that matter.
Treats lean as short and cuts useful commandsLean means only stable, high-impact facts. Keep build and test commands and short examples, and move detail into scoped files or linked docs.
Trims once and never revisits the fileRules go stale as the code changes. Say when you review it, such as after repeated agent mistakes and after architecture changes.
Interview lens
Likely follow-ups, what you can say, and the weak answers to avoid.
