The five rules

Pendraic auto-injects a lot of context around your prompt at run time: the project's logline, voice print, POV, tense, genre rules, surrounding scene. Your prompt body is just the instruction layer. These five rules keep that instruction tight enough to actually steer the output.

1.

One job per prompt

If you ask for “summarize and rewrite,” the model will do one of them well and the other badly. Make it two prompts. The Optimize button enforces this, drafts that try to do two things will come back split.

2.

Imperative phrasing

Start with a verb. “Tighten the selection” works. “The selection should be tighter” doesn't, it tells the model the goal but not who's doing the tightening.

3.

Under ~800 characters

Long prompts drown out the instruction. The auto-injected context already carries the project's rules, don't restate them. If your prompt is over 800 chars, hit Optimize.

4.

Constrain the output

“Return only the rewritten prose” suppresses the model's instinct to preface with “Here's a tighter version:” or wrap the answer in markdown. Every generation prompt should end with an explicit return rule.

5.

Smallest scope token

Use {{selection}} for paragraph edits. {{scene_text}} for scene-wide. {{synopsis_5p}} for whole-book. Match the scope to the job; don't pay for tokens you won't use.

Before / after

Before · 312 chars

I want you to please rewrite the selection I have highlighted, making it sound more punchy and less wordy and removing weak words and fillers, but please keep the meaning the same and don't add anything new. The voice should be the same.

After · 142 chars

Tighten {{selection}}. Cut filler and weak intensifiers. Keep meaning. Stay in {{pov}} {{tense}}. Return only the rewritten prose.

The Optimize button on the in-editor prompt editor runs this exact compression for you. Use it after you've drafted a custom prompt if you want a final pass.